Skip to content

CHANGELOG.md

Latest commit

 

History

History
1306 lines (1111 loc) · 125 KB

CHANGELOG.md

File metadata and controls

1306 lines (1111 loc) · 125 KB
Raw

Changelog

Unreleased

  • docs(adapters): auto-generated tier badge on every docs/adapters/*.md page, sourced from ADAPTER_REGISTRY.json. Closes LONG_TERM_FIXES_TRACKER T3.2. New script scripts/refresh_adapter_doc_badges.py rewrites a machine-fenced block (<!-- adapter-badge:begin ... :end -->) on each adapter doc; only text inside the fence is touched, everything else is preserved exactly. The badge summarizes tier (🟢 Core / 🟡 Extended), group, verbs (with truncation for long lists), privilege_tier, sandbox_safe, and network_facing. 11 adapter doc pages stamped: POSTGRES, MYSQL, REDIS, DYNAMODB, AIRTABLE, SUPABASE, AINL_GRAPH_MEMORY, MEMORY_CONTRACT, MEMORY_CONTRACT_V1_1_RFC, CODE_CONTEXT, PTC_RUNNER. Non-adapter docs are skipped via an explicit SKIP_FILES list. CI hook: --check mode exits 1 on drift; tests/test_adapter_doc_badges.py enforces the contract with 6 tests covering presence, skip-list integrity, registry-tier parity, idempotency, and drift detection. Side-effect honesty: the script also lists 16 Extended-tier adapters that don't have a dedicated doc page (solana, tiktok, etc.); informational, doesn't block.
  • docs(icp): canonical ICP doc at docs/WHO_IS_THIS_FOR.md. Closes LONG_TERM_FIXES_TRACKER T3.8. Synthesizes the three-baseline framework (LLM-first / hand-optimized / pure-deterministic), the persona table (solo operator / agent-heavy shop / small team needing audit / Temporal-curious / compliance-led / researcher), the four workload-pattern anti-fits (pure-deterministic automation, judgment-heavy semantic pipelines, already-gated triage, low-volume classifiers), the "what AINL is still good for when tokens aren't the play" table, a 30-second decision tree, and the reviewer Q&A into one source of truth. The README's 60-second filter is the teaser; this is the long-form answer. docs/competitive/WHEN_AINL_DOES_NOT_HELP.md is now a pointer stub (URL preserved for inbound links and search-engine results; full content merged into the canonical). docs/competitive/ARMARAOS_GTM.md retains its ArmaraOS-specific product positioning and gains a top banner pointing at the canonical for the ICP question. docs/AUDIENCE_GUIDE.md gains a top banner clarifying it's a contributor entry-point map, not an ICP filter. Sales-facing inbound links updated to the canonical across README, BENCHMARK, docs/benchmarks, CLAIMS_AND_EVIDENCE, competitive/README, competitive/COMPETITIVE_MESSAGING, competitive/OVERVIEW, competitive/COMPARISON_TABLE. The stub handles any unedited inbound link.
  • docs(token / community): transparent restatement of $AINL token + marketplace marketing surfaces. Closes LONG_TERM_FIXES_TRACKER T3.5. The principle: transparency beats revisionism. Aspirational marketing claims are preserved (git history is public; deletion would itself be a credibility risk), but every claim is now explicitly tagged with current shipping status and a pointer to live, verifiable data. New machine-readable canonical source: tooling/token_status.json (mint address, decimals, Solana on-chain placeholders, git-verifiable contributor counts, original aspirational goals retained as a fixed baseline). New refresher script: scripts/fetch_token_status.py — stdlib-only, no auth required for public reads, idempotent, populates on-chain (Solana RPC) + GitHub stars + local git contributor metric with explicit _status fields on every value (ok / pending_run / rpc_error: ... / github_error: ... / skipped); the script never invents numbers. New root CONTRIBUTORS.md with C3 framing — founder-led, multi-contributor, AI-assisted; active human maintainers: Steven B Hooley (founder), Terrance Schonleber, Kobe Welker; external human contributors and AI co-author identities listed; classification methodology (BOT_PATTERNS, AI_PATTERNS, HUMAN_ALIASES) documented and editable in the refresher script. Status-tagged: docs/community/CHAMPIONS_PROGRAM.md, docs/learning/intermediate/patterns/template-marketplace-submission.md, docs/learning/intermediate/patterns/README.md, docs/learning/intermediate/README.md, docs/learning/basics/README.md, docs/learning/basics/04-next-steps.md, docs/enterprise/ENTERPRISE_FAQ.md (per-utility status table), docs/enterprise/METRICS_DASHBOARD_SPEC.md (Token-holders metric clarified as community metric, not enterprise health signal). Enterprise stays token-free.
  • docs(adapters): two-tier adapter catalog — Core (29 adapters) + Extended (16 adapters) — both fully supported. No deprecations, no removals, no breaking changes. adapters/solana.py → adapters/extended/solana.py and adapters/tiktok.py → adapters/extended/tiktok.py; thin PEP 562 alias modules at the original paths forward every attribute (public and private) to the canonical extended modules so both from adapters.solana import ... and from adapters.extended.solana import ... continue to work identically and produce no warnings. Tier metadata stamped on every entry in ADAPTER_REGISTRY.json (tier: "core" | "extended", plus a new top-level tier_meanings block); regenerator + audit-driven TIER_MAP in scripts/update_adapter_registry_tiers.py. ainl doctor surfaces tier counts (adapter_tiers: 29 core + 16 extended (all supported)). New canonical doc docs/adapters/ADAPTER_TIERS.md describes the model, lists each tier, documents the Extended-adapter layered config pattern (explicit arg → env var → legacy default with one-time UserWarning → fail-fast), and records downstream compatibility commitments (ArmaraOS, ainl-inference-server, ainl-cortex, ainativelangweb). The Extended TiktokAdapter now uses that layered config and raises a clear TiktokAdapterConfigError when nothing is configured, replacing the previously hardcoded /Users/... path. Tests: tests/test_adapter_tier_shim_compat.py (9 cases — both import paths return the same class, no warnings on alias import, every adapter has a valid tier, layered config precedence, fail-fast). Tracker: docs/competitive/LONG_TERM_FIXES_TRACKER.md T3.1–T3.4 → Done.
  • docs(architecture): TOOL_SURFACE_CANONICALIZATION_CHECKLIST.md §9 — cross-repo vocabulary table (pre-filter vs D-gate vs router; snap-back vs future sticky); links to TOOL_REGISTRY_AND_TOOL_PATCH_PLAN.md §7.2–7.3 and ainl-inference-server (TOOL_PATCH_AND_ROUTER_ARCHITECTURE, PLANNER_HOST_BOUNDARIES, INFER_SESSION_ACTIVE_TODOS §N). Related rows in TOOL_REGISTRY_AND_TOOL_PATCH_PLAN.md §10 Related.
  • feat(mcp / cli): agent-recovery for empty / missing .ainl source. mcp_ainl_ainl_validate, mcp_ainl_ainl_compile, mcp_ainl_ainl_run, and mcp_ainl_ainl_security_report now short-circuit when code is whitespace-only with error_kind: "empty_source", tool_call_error: false, a minimal_strict_valid_example, and recommended_next_tools: ["ainl_get_started", "ainl_step_examples"] — distinct from the existing tool-wiring tool_call_error: true payload that fires when no code was passed at all (helpers _classify_source_arg / _empty_source_tool_error in scripts/ainl_mcp_server.py). The CLI runner (ainl run <file>) gains a pre-flight stat: error_kind: "empty_source" for 0-byte / whitespace-only files and error_kind: "path_not_found" (with a sibling_workflows listing of nearby .ainl / .lang files) when the path does not exist; both render as JSON under --json and as a short human summary otherwise (helpers _preflight_workflow_path / _pretty_preflight_error in cli/main.py). Tests: tests/test_mcp_server.py (new empty_source cases under TestValidate / TestCompile / TestRun), tests/test_cli_run_preflight.py. Docs: AGENTS.md (Empty / missing source self-recovery).
  • feat(mcp): path on ainl_validate, ainl_compile, ainl_run, and ainl_security_report — load workflow source from a filesystem path (inline code/ainl still takes precedence); optional env AINL_MCP_WORKFLOW_ROOT for resolving relative paths (ArmaraOS / multi-root workspaces). Mirrors CLI preflight (path_not_found, empty_source, path_not_a_file, siblings); success payloads may include source_path. ainl_run: workflows under an intelligence/ path segment set AINL_ALLOW_IR_DECLARED_ADAPTERS=1 when unset, matching CLI. Tests: tests/test_mcp_server.py. ArmaraOS host wiring: the OpenFang kernel additionally augments cached MCP tool definitions with the path property when an older ainl-mcp tools/list omitted it (openfang_types::tool::augment_mcp_ainl_tool_definition, applied at every MCP cache build site in openfang-kernel::kernel) so the planner schemas and chat tool lists carry path even before operators redeploy the Python server — see armaraos CHANGELOG.md (AINL MCP tool schemas — optional path) and CLAUDE.md (Common Gotchas).

v1.8.0 (April 26, 2026) — MCP authoring wizard, strict-valid corpus index, HTTP machine payments

  • feat(mcp / authoring): deterministic wizard in tooling/ainl_get_started.py with wizard_state_json continuity; new MCP tool ainl_step_examples; resource ainl://strict-valid-families backed by corpus/strict_valid_family_index.json (tooling/corpus_mining.py); ainl_validate / ainl_compile attach contract_validation_status and contract_alignment (mismatched_calls vs ADAPTER_CONTRACTS). Tests: tests/test_ainl_get_started.py, tests/test_mcp_server.py, tests/test_mcp_exposure_scoping.py.
  • feat(http): optional payment_profile (none/auto/x402/mpp) on SimpleHttpAdapter — structured 402 payment_required envelope + http_payment frame merges; settlement metadata on 200 when enabled; helpers in runtime/adapters/http_machine_payments.py. CLI --http-payment-profile, --http-max-payment-rounds; runner + MCP adapters.http keys; AINL_HTTP_PAYMENT_PROFILE / AINL_HTTP_MAX_PAYMENT_ROUNDS for OpenClaw monitor registry. Tests: tests/test_http_machine_payments.py, extended tests/test_http_adapter_contracts.py.
  • docs(integrations): docs/integrations/README.md (index); HTTP_MACHINE_PAYMENTS.md (x402/MPP + landscape + repo map); AGENTIC_PROTOCOLS_PRACTITIONER_READINESS.md (capability matrix + backlog); AGTP.md (gateway / MCP / adapter options); cross-links from AGENTS.md, README.md, SECURITY.md, examples/README.md, docs/EXAMPLE_SUPPORT_MATRIX.md.
  • docs: docs/operations/MCP_AINL_WIZARD_AND_CORPUS.md (hub); cross-links from AGENTS.md, docs/README.md, docs/DOCS_INDEX.md, examples/README.md, armaraos/docs/mcp-a2a.md, armaraos/docs/graph-memory.md; tooling/mcp_exposure_profiles.json — ainl_step_examples + ainl://strict-valid-families on design_impact_first, inspect_only, safe_workflow, full; validate_only includes ainl_step_examples without MCP resources.
  • test: refresh test_patch_a_valid_artifacts_fingerprint_lock golden hashes for opcode S core web fixtures (labels_sha256, emit_ir_json_sha256) after compiler IR shape drift — restores parser-compat / core-pr CI (open Dependabot PRs were failing the same gate).
  • test / tooling: drop duplicate a2a.GET / a2a.TASK_GET entries from ADAPTER_EFFECT (manifest verbs are discover, send, get_task only); align safe_workflow / full MCP exposure tool/resource lists with ALL_TOOL_NAMES / ALL_RESOURCE_URIS (includes ainl_step_examples); register 16 new examples/** paths in tooling/artifact_profiles.json (non-strict-only / legacy-compat so canonical curriculum stays unchanged); test_memory_search_missing_adapter_raises_descriptive uses host_adapter_denylist so auto AINLGraphMemoryBridge registration does not mask the missing-adapter path.

v1.7.1 (April 22, 2026) — A2A (Agent-to-Agent) adapter

  • feat(adapter / a2a): opt-in runtime/adapters/a2a.py — wire profile 1.0: GET …/.well-known/agent.json, JSON-RPC tasks/send / tasks/get; host allowlist (allow_hosts, empty list denies all), allow_insecure_local, optional strict_ssrf (DNS + block private/loopback/link-local unless local allowed), follow_redirects default off (on: re-check URL per hop). CLI: --a2a-allow-hosts, --a2a-allow-insecure-local, --a2a-strict-ssrf, --a2a-follow-redirects, --a2a-default-timeout, --enable-adapter a2a. Env: AINL_A2A_*, AINL_ADAPTERS. MCP / runner: adapters.enable + adapters.a2a; exposure profiles do not enable a2a by themselves — see tooling/mcp_exposure_profiles.json. Effect analysis: strict a2a.* in tooling/effect_analysis.py. Tests: tests/test_a2a_adapter.py, tests/test_a2a_adapter_integration.py (local HTTPServer; skips if bind fails). Example: examples/compact/a2a_delegate.ainl. Docs: docs/integrations/A2A_ADAPTER.md, ADAPTER_REGISTRY.json, registry/reference docs, SECURITY.md.

Known limitations (triage; full detail in A2A_ADAPTER)

  • TOCTOU — allowlist and DNS checks apply per request; targets can change between metadata fetch and work requests in adversarial conditions.
  • Empty allowlist — outbound a2a calls are denied; operators must set hosts explicitly.
  • IDNA / Unicode hosts — punycode and homoglyph risk; prefer explicit allowlisted hostnames.

[1.7.0] — ArmaraOS graph bridge + cognitive vitals (2026-04-14)

Ship tag for everything after v1.6.0 (7b78f18) through Gap K vitals parity: graph-memory inbox + monitor registry, Hand schema_version, Rust snapshot / export alignment, bundle pre-seed, OpenClaw wrapper fixes, and PRIOR_ART / blog operator docs. Full conventional list under v1.7.0 below.

  • feat(armaraos): inbox sync toward ainl_memory.db; monitor registry pre-seed + AdapterRegistry.get + unified bootstrap (CronDriftCheckAdapter); tests/test_armaraos_monitor_registry.py.
  • feat(armaraos-bridge): inbox JSON schema, CI workflow, sync envelope metadata (f1c4e62).
  • feat(armaraos-emit): schema_version on Hand artifacts (0bb96b8); Python bridge accepts Rust snapshot schema_version "1" (**a2a2c78`).
  • feat(ainl_graph_memory): read AgentGraphSnapshot via export env (f2e6372); .ainlbundle boot pre-seed for non-persona graph nodes (1e86f14); feat(runtime+bundle) pre-seed graph store on ainl run + persona hook (eece265).
  • fix(bridge): ArmaraOS per-agent export paths aligned (5a88026).
  • feat(gap-k): CognitiveVitals on episodic MemoryNode, inbox schema, tests/test_vitals_bridge.py (3af546c).
  • integration(armaraos-rust-crates): clarify production convergence across ainl-runtime, openfang-runtime, openfang-kernel, and openfang-types (optional ainl-runtime-engine turn routing, depth guards, shared graph-memory semantics) to reflect the shipped ArmaraOS bridge path.
  • integration(graphpatch/registry): document cross-runtime GraphPatch alignment — Python memory.patch remains the rich executor while Rust-side PatchAdapter / AdapterRegistry + GraphPatchAdapter host dispatch form the stable patch-registry contract.
  • integration(persona/extractor/tagger): align release notes with ArmaraOS runtime feature stack (ainl-persona-evolution, ainl-extractor, ainl-tagger) and runtime gates (AINL_PERSONA_EVOLUTION, AINL_EXTRACTOR_ENABLED, AINL_TAGGER_ENABLED) used for persona evolution, extraction reporting, and semantic tagging.
  • fix(openclaw): token_aware_startup_context includes + fs read (6a3b35b).
  • docs / chore: PRIOR_ART + graph-memory blog + inbox/bridge hub sync + bundle boot + OpenClaw cron graphs + supervisor_fixed.ainl; demo test_openspace_mcp HTTP MCP (c0a87c4); post-evolution ARMARAOS export refresh (c03d897); docs(release): v1.6.0 follow-up (8dcf043).

v1.7.0 (April 14, 2026) — Cognitive vitals on graph bridge + Hand schema_version + monitor registry

All items below are commits on main since Git tag v1.6.0 (7b78f18 — release: v1.6.0 — GraphPatch notes, version surfaces, adapter registry), through 3af546c, unless noted.

  • release: bump pyproject.toml / RUNTIME_VERSION / CITATION.cff / tooling/bot_bootstrap.json to 1.7.0 (mirrored tests/emits/server/runtime/engine.py); docs hub / skills / operations PyPI + docs pointers aligned for the v1.7.0 line (MINOR — bridge + inbox + vitals + emit surface).
  • feat(runtime+bundle) (eece265): pre-seed graph store on ainl run when a bundle / graph-memory path is active; persona hook wiring in the agent loop path (bundle + graph continuity).
  • docs (e3449fa, 8492a8a): ArmaraOS bundle env (AINL_BUNDLE_PATH, AINL_AGENT_ID), boot pre-seed, and chat persona vs JSON graph cross-links.
  • docs(release) (8dcf043): v1.6.0 GraphPatch release doc pass (changelog / release notes alignment — landed immediately after the v1.6.0 tag; included here for a complete post-tag audit trail).
  • fix(openclaw) (6a3b35b): intelligence/token_aware_startup_context.ainl — correct include paths and fs read wiring for bridge cron graphs.
  • feat(ainl_graph_memory) (f2e6372): read Rust AgentGraphSnapshot via AINL_GRAPH_MEMORY_ARMARAOS_EXPORT (and related export env) for Python-side import.
  • chore(demo) (c0a87c4): point demo/test_openspace_mcp.ainl at HTTP MCP execute_task (demo hygiene).
  • docs(ainl_graph_memory) (c03d897): document post-evolution ARMARAOS graph export refresh semantics.
  • fix(armaraos) (a2a2c78): accept Rust graph snapshot schema_version: "1" in the Python bridge (forward-compat with emitted packs).
  • feat(armaraos-emit) (0bb96b8): emit schema_version = "1" on HAND.toml [hand], shallow-copied IR JSON, and security.json (openfang-hands alignment).
  • fix(bridge) (5a88026): align ainl_graph_memory ArmaraOS export discovery with per-agent {agent}_graph_export.json / agent directory layout.
  • feat(ainl_graph_memory) (1e86f14): pre-seed non-persona graph nodes from .ainlbundle on boot() (bundle round-trip toward JSON graph store).
  • docs(prior-art) (04dbbe2): expand graph-memory timeline + implementation claims in PRIOR_ART.md.
  • feat(armaraos) (1300979): inbox sync for graph-memory mutations toward ainl_memory.db (Python writer + host drain contract).
  • feat(armaraos) (4ad914c): pre-seed monitor registry with bridge adapters (cron drift + graph tooling path).
  • feat(armaraos) (1092daa): unify monitor registry bootstrap + expose AdapterRegistry.get(name) publicly; RuntimeEngine graph-patch paths resolve ainl_graph_memory via adapters.get(...) (no private _adapters access).
  • feat(armaraos-bridge) (f1c4e62): inbox JSON schema (ainl_graph_memory_inbox_schema_v1.json), CI workflow coverage, sync envelope metadata (source_features, etc.).
  • docs (965b8d5, 843be1a, 1ed8ab2, a7dfb65, eeeede8): graph-memory inbox contract + vendored armaraos hub sync + bundle graph boot + adapters index + OpenClaw wrapper include / fs / If notes for bridge cron graphs + graph-as-memory blog / PRIOR_ART / adapter frontmatter cross-links.
  • feat(gap-k) (3af546c): CognitiveVitals round-trip on Python MemoryNode (vitals_gate, vitals_phase, vitals_trust); from_dict / to_dict; Rust snapshot import; inbox schema + tests/test_vitals_bridge.py.
  • chore: scripts/wrappers/supervisor_fixed.ainl — canonical R openclaw_memory append_today adapter form (wrapper hygiene).

[1.6.0] — GraphPatch complete (2026-04-12)

Ship tag for the GraphPatch line; full conventional list under v1.6.0 below.

  • GraphPatch op end-to-end: bridge (graph_patch), runtime (install + fitness + boot reinstall), compiler (strict literals / dataflow), tooling (MEMORY_PATCH / effect analysis), tests (tests/test_graph_patch_op.py).
  • Guards / strictness: OverwriteGuardError on compiled-label collision; StrictModeError for invalid memory.patch literals when strict_literals is enabled.
  • Fitness + boot: fitness EMA on patched-label exit (including early J); _reinstall_patches reapplies active patches after engine boot when the graph-memory bridge is present.
  • Dispatch: R memory.patch is routed through the same engine memory adapter dispatch path as other memory.* verbs (not a special-case bypass).
  • Foundation for unified graph execution: this release locked in canonical Python GraphPatch semantics and dataflow guarantees that the ArmaraOS Rust integration extends in v1.7.0 via patch-registry dispatch and graph-memory convergence.

v1.6.0 (April 12, 2026) — GraphPatch (memory.patch), strict dataflow, bridge graph_patch

  • release: bump pyproject.toml / RUNTIME_VERSION / CITATION.cff / tooling/bot_bootstrap.json to 1.6.0 (mirrored tests/emits/server/runtime/engine.py).
  • feat(runtime / graph memory): R memory.patch installs procedural label bodies from ainl_graph_memory via adapters.call("ainl_graph_memory", "graph_patch", [memory_node_id, label_name], …); _reinstall_patches on engine boot; overwrite guard for compiled labels (OverwriteGuardError → AinlRuntimeError); patch metadata (__patch_node_id__, __patch_version__, __fitness__) + finalize_patch declared reads; fitness EMA on label exit including early J returns; sync/async memory dispatch includes patch.
  • feat(compiler): strict-mode memory.patch literal guard (StrictModeError / strict_literals on AICodeCompiler); patch dataflow validation uses per-step _analyze_step_rw (aligns with compiler read analysis for string Set refs).
  • feat(tooling): ainl_graph_memory.MEMORY_PATCH in ADAPTER_EFFECT (tooling/effect_analysis.py).
  • fix(runtime): Loop/While inner locals renamed (body_lid) so inner graphs do not shadow body (patch fitness updates).
  • test: tests/test_graph_patch_op.py — eight pytest cases (step vs graph mode, frame $var, overwrite guard, re-patch versioning, boot reinstall + _reinstall_patches, fitness EMA, strict compile).
  • docs: docs/RELEASE_NOTES.md, docs/POST_RELEASE_ROADMAP.md, docs/RELEASING.md, ADAPTER_REGISTRY.json / docs/reference/ADAPTER_REGISTRY.md (memory_patch target for GraphPatch; runtime call target graph_patch), current-release pointers to v1.6.0 across hub and integration docs.
  • docs: docs/adapters/AINL_GRAPH_MEMORY.md — boot(), AINL_BUNDLE_PATH / AINL_AGENT_ID, scheduled ainl run bundle round-trip vs Rust ainl_memory.db chat [Persona traits active: …] hook; docs/ARMARAOS_INTEGRATION.md — env table rows + cron / chat split + armaraos docs/graph-memory.md; docs/INTELLIGENCE_PROGRAMS.md — stateful bundle.ainlbundle + armaraos doc links; AGENTS.md (ArmaraOS bullets).

v1.5.2 (April 12, 2026) — Graph memory IR substrate, bundle serialization, operator surfaces

  • release: bump pyproject.toml / RUNTIME_VERSION / CITATION.cff / tooling/bot_bootstrap.json to 1.5.2 (mirrored tests/emits/server/runtime/engine.py).
  • feat(memory / graph IR): optional memory_type on per-label R nodes for graph-memory / persona steps; memory.pattern_recall on the graph bridge + __last_pattern__ for memory.merge; per-label emit_edges (port: "data" / "emit") alongside required_emit_targets; tooling/graph_api helpers emit_edges, data_flow_edges, memory_nodes (label_* accept label id sequences). Docs: GRAPH_SCHEMA.md, GRAPH_INTROSPECTION.md, AINL_GRAPH_MEMORY.md, MEMORY_CONTRACT.md.
  • feat(compiler / runtime / bundle): graph-memory and persona adapter ops registered for strict validation (OP_REGISTRY, MODULE_ALIASES, ADAPTER_EFFECT, KNOWN_MODULES as applicable); persona.load with runtime frame injection (__persona__, persona_instruction); AINLBundle / AINLBundleBuilder in runtime/ainl_bundle.py for single .ainlbundle workflow + memory + persona + tools JSON round-trip.
  • feat(mcp): malformed workspace ainl_mcp_limits.json → successful ainl_run may include warnings; invalid non-empty workspace cache.json during MCP auto-registration → adapter_config_error with details.
  • feat(emit/armaraos): security.json includes capability_declarations.adapters; HAND.toml [hand] includes ainl_ir_version.
  • fix(runtime): enforce max_adapter_calls including 0 (no longer folded to “unset”); aligns with MCP/workspace limits and tests.
  • fix(profiles): local_minimal max_adapter_calls raised from 0 to 500 in tooling/security_profiles.json so the profile remains usable with the stricter adapter-call ceiling.
  • test: tests/test_mcp_frame_hints.py, tests/test_mcp_workspace_limits.py, tests/test_mcp_auto_cache_adapter.py, tests/test_emit_armaraos_handpack.py, tests/test_compact_opcode_ir_parity.py, tests/test_memory_search_op.py, tests/test_core_builtins_v143.py, persona / bundle / strict-contract suites — see docs/RELEASE_NOTES.md.
  • test(conftest): offline_llm_provider_config pytest fixture (deterministic offline LLM provider block for suites that opt in via pytestmark).
  • docs: docs/adapters/MEMORY_CONTRACT.md, docs/adapters/AINL_GRAPH_MEMORY.md, docs/architecture/GRAPH_INTROSPECTION.md, docs/RUNTIME_COMPILER_CONTRACT.md, docs/DOCS_INDEX.md, docs/adapters/README.md, docs/architecture/STATE_DISCIPLINE.md, docs/README.md, docs/RELEASE_NOTES.md, AGENTS.md, README.md, docs/operations/MCP_RESEARCH_CONTRACT.md, docs/ARMARAOS_INTEGRATION.md, docs/operations/CAPABILITY_GRANT_MODEL.md, hub / overview / skills — current-release pointers aligned to v1.5.2; WHITEPAPERDRAFT.md (v1.5.2, §6.8 graph-memory gap audit).

v1.5.1 (April 12, 2026) — Graph memory runtime + ArmaraOS bridge docs

  • release: bump pyproject.toml / RUNTIME_VERSION / CITATION.cff / tooling/bot_bootstrap.json to 1.5.1 (mirrored tests/emits/server/runtime/engine.py).
  • feat(runtime): IR ops MemoryRecall and MemorySearch dispatch the ainl_graph_memory adapter from _exec_step and sync _run_label_graph; fallback adapter inference includes legacy steps and label graph nodes (_fallback_adapters_from_label_steps). Tests: tests/test_memory_recall_op.py.
  • bridge (ArmaraOS): JSON graph store + AINLGraphMemoryBridge (armaraos/bridge/ainl_graph_memory.py), registration and on_delegation hook in armaraos/bridge/runner.py, optional armaraos/bridge/graph_viz/ FastAPI viewer, demo/procedural_roundtrip_demo.py, demo/ainl_graph_memory_demo.py (stdlib walkthrough of episodic / semantic / procedural / persona nodes + JSON export; run from repo root), armaraos/bridge/bridge_token_budget_adapter.py importlib shim.
  • docs: new docs/adapters/AINL_GRAPH_MEMORY.md; cross-links from docs/adapters/MEMORY_CONTRACT.md, docs/adapters/README.md, docs/README.md, docs/DOCS_INDEX.md, docs/reference/README.md, docs/ARMARAOS_INTEGRATION.md ( AINL_GRAPH_MEMORY_PATH ), docs/RUNTIME_COMPILER_CONTRACT.md, docs/architecture/STATE_DISCIPLINE.md; hub / overview / skills / README.md “current release” lines aligned to v1.5.1.
  • tooling(docs): register ainl_graph_memory in tooling/adapter_manifest.json and ADAPTER_EFFECT (tooling/effect_analysis.py) with graph-node effect tiers; extend docs/reference/ADAPTER_REGISTRY.md, ADAPTER_REGISTRY.json, docs/GITHUB_RELEASE_CHECKLIST.md, docs/DOCS_MAINTENANCE.md, and docs/CONFORMANCE.md (strict-literals matrix callout).

v1.5.0 (April 10, 2026) — Minor release + documentation alignment

  • release: bump pyproject.toml / RUNTIME_VERSION / CITATION.cff / tooling/bot_bootstrap.json to 1.5.0 (minor after 1.4.6).
  • docs(whitepaper): WHITEPAPERDRAFT.md — substantive pass for 1.5.0: runner default grants vs ainl_run adapter registration on MCP, host_security_env / capability-block metrics, ainl-validate hyperspace path, adapter and emission accuracy, include prelude; new §16–§18 (security / CI & assurance / LLM authoring contract); renumbered tail (§19–§22); §20.1 shipped-features parity; Appendix A pointers to docs/advanced/SAFE_USE_AND_THREAT_MODEL.md and AGENTS.md.
  • docs: repository-wide refresh — docs hub (docs/README.md), docs/DOCS_INDEX.md, docs/overview/README.md, docs/OPENCLAW_INTEGRATION.md, docs/CONTRIBUTING.md, root CONTRIBUTING.md, docs/operations/OPENCLAW_AINL_GOLD_STANDARD.md / docs/operations/README.md / docs/operations/HOST_PACK_OPENCLAW.md / docs/operations/EXTERNAL_ORCHESTRATION_GUIDE.md / docs/operations/OPENCLAW_REQUESTS_2_6_MAPPING.md, docs/competitive/README.md / docs/competitive/OVERVIEW.md, docs/openclaw/TOKEN_AWARE_STARTUP_CONTEXT.md, docs/POST_RELEASE_ROADMAP.md, docs/upstream/OPENCLAW_ISSUE_DRAFT_bootstrap_flag.md, docs/operations/MCP_RESEARCH_CONTRACT.md, docs/RELEASING.md, docs/QUICKSTART_OPENCLAW.md, docs/getting_started/HOST_MCP_INTEGRATIONS.md, docs/ZEROCLAW_INTEGRATION.md, docs/HERMES_INTEGRATION.md, docs/integrations/hermes-agent.md, docs/solana_quickstart.md, README.md, AGENTS.md, WHITEPAPERDRAFT.md — “current release” / PyPI pointers (and JSON samples where applicable) aligned to 1.5.0; skills/openclaw/README.md / SKILL.md, skills/ainl/README.md / SKILL.md, skills/hermes/README.md / SKILL.md — PyPI / host-briefing lines aligned to v1.5.0.

v1.4.6 (April 11, 2026) — Workspace samples + OpenSpace dev harness

  • release: bump pyproject.toml / RUNTIME_VERSION / CITATION.cff / tooling/bot_bootstrap.json to 1.4.6.
  • chore(demo): add demo/test_openspace_http.ainl — experimental HTTP probe for OpenSpace MCP (demo tree; not strict-valid).
  • chore: add apollo-x-bot/api-cost-monitor.ainl — sample promoter workflow for API-cost monitoring (uses shared modules/common includes).
  • chore: add run_openspace_test.py — portable dev harness for demo/test_openspace_mcp.ainl (repo-root-relative paths; OpenClaw adapter registry smoke).

v1.4.5 (April 10, 2026) — ArmaraOS MCP env merge, MCP authoring surface, compiler diagnostics

  • release: bump pyproject.toml / RUNTIME_VERSION / CITATION.cff / tooling/bot_bootstrap.json to 1.4.5.
  • fix(tooling): ainl install-mcp --host armaraos — when name = "ainl" already exists in [[mcp_servers]], merge env pass-through (AINL_MCP_EXPOSURE_PROFILE, AINL_MCP_TOOLS, AINL_MCP_TOOLS_EXCLUDE, AINL_MCP_RESOURCES, AINL_MCP_RESOURCES_EXCLUDE) into the existing env = [...] line (sorted union) instead of only applying to newly appended blocks. ~/.openfang/config.toml receives the same merge. Tests in tests/test_install_mcp.py.
  • feat(mcp): scripts/ainl_mcp_server.py — register ainl://authoring-cheatsheet resource (short golden-path authoring guide aligned with AGENTS.md); process-local validate telemetry counters; validate/compile path improvements for MCP agent loops.
  • feat(compiler / diagnostics): Diagnostic supports optional contract_violation_reason for include failures; include diagnostics populate stable reasons; graph validation emits better line attribution (_graph_error_lineno, _emit_graph_validation_diagnostic).
  • fix(compiler strict): Targeted labels whose last step is Loop or While are skipped for the “exactly one J / must end in J” check (same exemption pattern as If-final labels). Fixes false strict failures for assembly-style pipelines.
  • test: tests/test_compiler_agent_suggested_fixes.py — coverage for suggested-fix kinds on common agent mistakes; tests/test_mcp_server.py extended for MCP surface changes.
  • docs: docs/operations/EFFICIENT_MODE_ARMARAOS_BRIDGE.md — clarifies ainl run --efficient-mode / AINL_EFFICIENT_MODE (env signal only), modules/efficient_styles.ainl (output density), vs ArmaraOS Rust input compression (see ArmaraOS docs/prompt-compression-efficient-mode.md).

v1.4.4 (April 9, 2026) — PyPI packaging + discoverability alignment

  • release: bump pyproject.toml / RUNTIME_VERSION / CITATION.cff / tooling/bot_bootstrap.json to 1.4.4.
  • fix(emit): emit_solana_client discoverability header interpolates RUNTIME_VERSION (was hard-coded v1.4.2).
  • test: Solana discoverability assertions use f"v{RUNTIME_VERSION}" instead of a pinned version string.

v1.4.3 (April 8, 2026) — MCP per-run adapter configuration + core builtins expansion

  • feat(mcp): ainl_run now accepts an optional adapters argument to enable scoped runtime adapters per call (sandboxed fs, host-allowlisted http, file-backed cache, optional sqlite) so agent workflows can do required I/O without asking end users to edit global config.
  • feat(builtins): CoreBuiltinAdapter (runtime/adapters/builtins.py) now implements the full set of verbs that were already present in the validator contract but missing at runtime: EQ, NEQ, GT, LT, GTE, LTE (comparisons); TRIM (collapse whitespace + strip), STRIP, LSTRIP, RSTRIP (edge whitespace); STARTSWITH, ENDSWITH (string predicates); KEYS, VALUES (dict introspection); STR, INT, FLOAT, BOOL (type coercions). All purely additive — no existing verb changed. These eliminate the common workarounds (CONCAT "" val for int→str coercion, cascading REPLACE for whitespace normalization, CONTAINS for equality checks).
  • feat(mcp): ainl_compile now returns frame_hints[] — a list of {name, type, source} entries describing variables the caller should supply via frame in ainl_run. Sources: explicit # frame: name: type comment lines (authoritative) and IR-inferred undeclared variables (heuristic).
  • feat(mcp): Per-workspace ainl_mcp_limits.json override — if <fs.root>/ainl_mcp_limits.json exists, its values are merged into caller limits before server ceiling enforcement. Enables workspace-specific limit tuning without editing global defaults.
  • feat(mcp): Auto-registered cache adapter — if fs is enabled and cache is not explicitly configured, the MCP server automatically registers cache when output/cache.json or cache.json exists in fs.root.
  • fix(limits): _SERVER_DEFAULT_LIMITS in scripts/runtime_runner_service.py raised to match MCP defaults (max_steps: 500000, max_depth: 500, max_adapter_calls: 50000, max_time_ms: 900000) — the two surfaces were inconsistent since the previous session's MCP-only raise.
  • fix(tooling): tooling/effect_analysis.py now includes core.STARTSWITH, core.ENDSWITH, core.TRIM, core.STRIP, core.LSTRIP, core.RSTRIP effect entries matching the new builtins.
  • fix(docs): ainl_run docstring documents variable shadowing pitfall and # frame: convention. CAPABILITY_GRANT_MODEL.md, RUNTIME_CONTAINER_GUIDE.md, SANDBOX_EXECUTION_PROFILE.md updated with current default limits. MCP_RESEARCH_CONTRACT.md updated for frame_hints, workspace limits, and auto-cache. AGENTS.md runtime verb list updated.
  • release: bump pyproject.toml / RUNTIME_VERSION / CITATION.cff / tooling/bot_bootstrap.json to 1.4.3.

v1.4.2 (April 7, 2026) — Intelligence path policy, MCP/runner alignment, compiler strict-mode + tooling

  • feat(runtime): AINL_ALLOW_IR_DECLARED_ADAPTERS — when set, AINL_HOST_ADAPTER_ALLOWLIST from the environment is ignored so graphs can use IR-declared adapters; denylist and security profiles still apply; ainl doctor / capability hints updated. For sources under an intelligence/ path segment, RuntimeEngine.from_code and ainl run set AINL_ALLOW_IR_DECLARED_ADAPTERS=1 when unset, unless AINL_INTELLIGENCE_FORCE_HOST_POLICY=1.
  • feat(cli): ainl run registers web, tiktok, queue (OpenClaw integration) consistently on every invocation.
  • feat(mcp): scripts/ainl_mcp_server.py — default ainl_run / fitness tool grants aligned with the HTTP runner (no core-only adapter cap at the policy layer; resource limits unchanged). LLM on MCP remains opt-in via AINL_CONFIG / AINL_MCP_LLM_ENABLED.
  • feat(runner): scripts/runtime_runner_service.py documents the adapter-relax flag in host_security_env / capabilities response.
  • test: tests/test_host_adapter_allowlist_env.py — intelligence path relax, host allowlist env behavior, strict policy override coverage.
  • docs: AGENTS.md, docs/INTELLIGENCE_PROGRAMS.md, docs/LLM_ADAPTER_USAGE.md, env/openclaw-workspace-ainl.env.example, ArmaraOS scheduled-ainl / snippet cross-links.
  • fix(compiler): J L_label (jump to a known label by name) no longer generates false "undefined variable" strict-mode errors. _steps_to_graph now emits analysis_only: true inter-label edges for every J node whose var resolves to a known label, enabling inter-label dataflow propagation without confusing the dataflow checker.
  • fix(compiler): Edge port validation now skips analysis_only: true edges (compiler-internal, dataflow-only), eliminating the spurious "edge … invalid for op 'J' (allowed: [])" error that appeared when strict mode validated label-jumping graphs.
  • fix(tooling): strict_adapter_key_for_step in tooling/effect_analysis.py now uses the IR entity field as the verb fallback when req_op is empty. This fixes *.F unknown-adapter-verb errors for R adapter verb args instructions whose IR stores the verb in entity (affects core, web, tiktok, cache, svc, crm, and similar adapters).
  • feat(tooling): Expanded ADAPTER_EFFECT allowlist with:
    • web.* — SEARCH, FETCH, SCRAPE, GET
    • tiktok.* — RECENT, SEARCH, PROFILE, STATS, TRENDING
    • svc.* — STATUS, RESTART, CADDY, NGINX, HEALTH
    • crm.* — QUERY, UPDATE
    • Additional core.* ops: LEN, LT, GT, EQ, MAP, FILTER, CONCAT, SLICE, PARSE, FORMAT, NOW, ISO, JOIN, KEYS, CONTAINS, MERGE, TYPE, BOOL, UPPER, LOWER, ROUND, ABS, MOD, FLOOR, CEIL, RANGE, ZIP, REDUCE, SORT, UNIQUE, FLAT, HEAD, TAIL, CHUNK, SAMPLE, SHUFFLE
  • feat(tooling): Updated tooling/adapter_manifest.json to include the web adapter namespace and all new verbs for tiktok, svc, crm, and core, keeping adapter manifest coverage tests green.
  • fix(intelligence): intelligence/intelligence_digest.lang — removed include "modules/common/generic_memory.ainl" as genmem (include was after the S header, violating strict prelude placement, and generic_memory.ainl does not conform to the ENTRY/EXIT_* module contract); inlined memory writes using direct R memory put adapter calls; replaced QueuePut notify note with R queue Put "notify" note ->_; added missing J L_analyze terminator to L_default_prev.
  • fix(intelligence): intelligence/infrastructure_watchdog.lang — replaced non-standard Ret null with J null; replaced QueuePut notify alert_msg with R queue Put "notify" alert_msg ->_.
  • feat(armaraos): ainl-library walker respects a .ainl-library-skip marker file — any directory tree containing this file is excluded from the App Store listing. Added demo/.ainl-library-skip so development-only demo files stay out of the user-facing App Store.
  • fix(profiles): Promoted 11 files from non-strict-only to strict-valid in tooling/artifact_profiles.json following resolution of their structural issues: examples/api_only.lang, examples/blog.lang, examples/ecom.lang, examples/internal_tool.lang, examples/ticketing.lang, examples/openclaw/daily_lead_summary.lang, examples/openclaw_full_unification.ainl, examples/test_if_var.ainl, examples/test_mul.ainl, examples/test_X_sub.ainl; added new examples/compact/openclaw_learning_handoff.ainl.
  • chore: Synced tooling/canonical_curriculum.json, tooling/canonical_training_pack.json (rebuilt via scripts/build_canonical_training_pack.py), and tests/fixtures/snapshots/compile_outputs.json with promoted files and refreshed checksums.
  • release: bump pyproject.toml / RUNTIME_VERSION / CITATION.cff / tooling/bot_bootstrap.json to 1.4.2.

v1.4.1 (April 3, 2026) — Wishlist CI, offline LLM provider, core.GET

  • feat(llm): register offline AbstractLLMAdapter (deterministic, no network) for config.yaml + register_llm_adapters demos and CI; LLMRuntimeAdapter normalizes verb casing so R llm.COMPLETION matches completion.
  • feat(core): implement core.GET on CoreBuiltinAdapter (deep key/index read via deep_get); add strict core.GET + llm.COMPLETION entries to tooling/effect_analysis.py.
  • examples(wishlist): add 05b_unified_llm_offline_config.ainl + fixtures/llm_offline.yaml — unified llm path vs 05_route_then_llm_mock.ainl (llm_query + mock env).
  • ci: parser-compat runs tests/test_wishlist_examples_strict.py plus no-network smoke runs for wishlist 01 and 05b.
  • release: bump pyproject.toml / RUNTIME_VERSION / CITATION.cff / tooling/bot_bootstrap.json to 1.4.1.

v1.4.0 (April 1, 2026) — ArmaraOS host pack + release readiness

  • release: bump pyproject.toml / PyPI ainl 1.4.0; align RUNTIME_VERSION 1.4.0 in runtime/engine.py (mirrored tests/emits/server/runtime/engine.py); CITATION.cff version / date-released; tooling/bot_bootstrap.json schema version; ainl serve GET /health reports version from RUNTIME_VERSION.
  • feat(armaraos): first-class ArmaraOS host-pack support:
    • ainl emit --target armaraos emits a hand package (HAND.toml, <stem>.ainl.json, security.json, README.md).
    • ainl status --host armaraos uses consistent env resolution (canonical ARMARAOS_*, legacy OPENFANG_* aliases).
    • ainl install-mcp --host armaraos supports ArmaraOS ~/.armaraos/config.toml format ([[mcp_servers]] array) and installs ~/.armaraos/bin/ainl-run + PATH hints.
  • docs(armaraos): new integration doc docs/ARMARAOS_INTEGRATION.md; linked from the host hub docs/getting_started/HOST_MCP_INTEGRATIONS.md and docs/DOCS_INDEX.md.
  • fix(tests): repair ArmaraOS integration tests to use supported AINL syntax and fix ArmaraOS emitter import path.

v1.3.3 (March 29, 2026) — PyYAML for ainl-mcp entrypoint

  • fix(packaging): add PyYAML to core dependencies. scripts/ainl_mcp_server.py imports yaml at module load; ainl-mcp --help / CI wheel-integrity must not fail with ModuleNotFoundError: yaml after pip install ainativelang[mcp].
  • fix(ci): tests/test_intelligence_budget_hydrate.py / tests/test_runtime_api_compat.py — close MemoryAdapter SQLite connections (and gc.collect) before deleting temp DB paths on Windows (avoids PermissionError in core-pr).
  • chore(benchmarks): refresh tooling/benchmark_size_ci.json after examples/hello.ainl grew (commented tutorial header); keeps benchmark-regression size gate aligned with minimal_emit token counts.

v1.3.2 (March 29, 2026) — Core HTTP deps for LLM adapter imports

  • fix(packaging): declare httpx and requests as core dependencies in pyproject.toml. adapters/__init__.py eagerly imports LLM adapters that require these modules; a bare pip install ainativelang[mcp] (or base install) must not fail with ModuleNotFoundError on ainl --help / import smoke (CI install-smoke / wheel-integrity).

v1.3.1 (March 29, 2026) — Native Solana + prediction markets (strict graphs), discoverability docs, mirrored runtime sync

  • release: bump pyproject.toml / PyPI ainl 1.3.1; RUNTIME_VERSION 1.3.1 in runtime/engine.py and tests/emits/server/runtime/engine.py; CITATION.cff version / date-released; tooling/bot_bootstrap.json schema version.
  • solana: prediction-market flows on adapters/solana.SolanaAdapter — DERIVE_PDA with single-quoted JSON seeds, GET_PYTH_PRICE (legacy + PriceUpdateV2), HERMES_FALLBACK, INVOKE / TRANSFER_SPL with priority fees; AINL_DRY_RUN=1 simulation envelopes for mutating verbs; optional pip install "ainativelang[solana]".
  • examples / curriculum: examples/solana_demo.ainl, examples/prediction_market_demo.ainl strict-valid; wired through artifact profiles / canonical curriculum / training packs where applicable.
  • emitters: --emit solana-client / blockchain-client standalone solana_client.py with v1.3.1 DISCOVERABILITY module docstring; see docs/emitters/README.md Solana section.
  • docs: docs/solana_quickstart.md, root README.md Solana callout, examples/README.md pointers, CONTRIBUTING.md (release version + Solana pointers), docs/CONTRIBUTING.md hub stub; tests assert discoverability strings in emitted client + adapter docstrings (tests/test_solana_blockchain.py).

v1.3.0 (March 27, 2026) — Official Hermes Agent integration + OpenClaw integration improvements (MCP bootstrap + skill pack + hermes-skill emitter)

  • solana / lexer: R solana.DERIVE_PDA seeds accept single-quoted JSON '["a","b"]' as one lexer token; under AINL_DRY_RUN=1, mutating verbs INVOKE, TRANSFER, and TRANSFER_SPL return simulated envelopes without requiring solders (live RPC/signing still needs ainativelang[solana]). Legacy tokenize_line() now delegates to tokenize_line_lossless (decoded slot values; same as compile).
  • code_context adapter: Added GET_SKELETON (cheap Tier-0 signatures), embedding-aware COMPRESS_CONTEXT (cosine ranking when embedding_memory is available), and extended STATS with graph metrics; plus import-graph GET_DEPENDENCIES / GET_IMPACT and greedy packing on earlier iterations. Builds on ctxzip tiers + forgeindex graph/compression ideas. See docs/adapters/CODE_CONTEXT.md, examples/code_context_demo.ainl. Full credit to Brady Drexler (ctxzip) and Chris Micah (forgeindex).
  • feat(hermes): add official Hermes Agent host support to ainl install-mcp --host hermes / ainl hermes-install (writes ~/.hermes/config.yaml mcp_servers.ainl, installs ~/.hermes/bin/ainl-run, PATH hint).
  • feat(skills/hermes): ship a Hermes-native skills pack under skills/hermes/ (installer + bridge helpers for ingest/export loops).
  • feat(emitter): add --emit hermes-skill (and --target hermes alias) to compile a .ainl workflow into a drop-in Hermes skill bundle (SKILL.md, workflow.ainl, ir.json) that runs deterministically via MCP ainl_run.
  • feat(doctor): ainl doctor now recognizes Hermes’ YAML host config (~/.hermes/config.yaml mcp_servers:) and validates ainl MCP registration without requiring a YAML dependency.
  • docs(hermes): new docs: docs/integrations/hermes-agent.md (high-level) and docs/HERMES_INTEGRATION.md (full guide, quickstarts, loop contract, troubleshooting).
  • docs(discovery): README badge + “Start here” + docs/README.md + docs/DOCS_INDEX.md + MCP host hub cross-link Hermes Agent next to OpenClaw / ZeroClaw so humans and agents can find the upstream host.
  • feat(openclaw): ainl install openclaw --workspace PATH for true one-command setup with health check, --dry-run support, and core gold-standard cron registration.
  • feat(openclaw): ainl status as a unified view for budget, cron health, and token usage (legacy weekly_remaining_v1 table plus memory_records fallback via _read_weekly_remaining_rollup).
  • feat(openclaw): self-healing validator and ainl doctor --ainl for OpenClaw + AINL integration checks.
  • fix(ux): improved error messages with actionable fix suggestions.
  • docs(openclaw): updated OpenClaw docs with progressive disclosure — quickstart-first, gold-standard depth preserved.
  • fix(openclaw): weekly budget display in ainl status correctly reads from modern memory_records primary storage (legacy table remains bootstrapped for compatibility).

v1.2.10 (March 27, 2026) — Wheel packaging fix, LLM adapter/monitoring pack, and release/docs sync

  • fix(packaging): include intelligence and intelligence.* in setuptools package discovery and add intelligence/__init__.py so wheel/PyPI installs include intelligence.signature_enforcer for CLI paths that import validator utilities.
  • fix(cli): ainl visualize now works from clean pip install ainativelang environments (no ModuleNotFoundError: intelligence during scripts.visualize_ainl import path).
  • feat(adapters/llm): add an internal LLM adapter layer under adapters/llm/ with AbstractLLMAdapter, LLMResponse/LLMUsage, and concrete OpenRouterAdapter and OllamaAdapter implementations registered via adapters/registry.AdapterRegistry; wiring is exercised in tests/test_adapters.py and documented in docs/ADAPTER_DEVELOPER_GUIDE.md.
  • feat(monitoring): introduce an optional AINL-native monitoring pack under intelligence/monitor/:
    • collector.MetricsCollector (in-memory metrics with Prometheus exposition),
    • cost_tracker.CostTracker (SQLite-backed LLM usage and budget store),
    • budget_policy.BudgetPolicy (monthly budget thresholds + Telegram alerts),
    • health.HealthStatus (liveness/readiness checks),
    • dashboard.server (Flask dashboard at /, /api/budget, /api/metrics, /health/* with a minimal static HTML UI). The pack is driven by Python-side intelligence programs and adapters, not the AINL language itself. Integration guides live in docs/MONITORING_OPERATIONS.md and docs/INTELLIGENCE_PROGRAMS_INTEGRATION.md.
  • feat(openclaw tools): add thin OpenClaw tool helpers under adapters/tools/openclaw/ and a simple MCP tools client/registry under adapters/tools/mcp/ to demonstrate how external tool surfaces can be wrapped for AINL-hosted agents; these are extension-level helpers, not canonical language features.
  • release: bump package/runtime surfaces to 1.2.10 (pyproject.toml, runtime/engine.py, tests/emits/server/runtime/engine.py, CITATION.cff).
  • docs: update release tracking docs to point at v1.2.10 and call out the PyPI quickstart flow (pip install → ainl init → ainl check → ainl run → ainl visualize). Cross-link the new monitoring/adapter components from README.md, WHITEPAPERDRAFT.md (§10.5/§13.5), docs/DOCS_INDEX.md, and the intelligence/operations docs so operators can discover the monitoring pack from both OpenClaw and core AINL entry points.

v1.2.9 (March 26, 2026) — PTC-Lisp hybrid integration polish (Phases 1–4.5)

  • feat(ptc): adapters/ptc_runner.py — opt-in PTC Runner adapter (HTTP + mock + subprocess mode via AINL_PTC_USE_SUBPROCESS / AINL_PTC_RUNNER_CMD); disabled by default; AINL_ENABLE_PTC=true or --enable-adapter ptc_runner; structured result envelope with ok, result, beam_metrics, beam_telemetry; health/status verbs; _strip_private_keys context firewall; registered in ADAPTER_REGISTRY.json, tooling/adapter_manifest.json, tooling/effect_analysis.py.
  • feat(ptc): adapters/llm_query.py — lightweight opt-in LLM query adapter mirroring ptc_runner structure; run/query verbs; _strip_private_keys at serialization boundary.
  • feat(ptc/modules): modules/common/ptc_run.ainl — thin syntactic-sugar wrapper over ptc_runner run with named frame variables and safe defaults for optional signature and subagent_budget.
  • feat(ptc/modules): modules/common/ptc_parallel.ainl — pcall-style fan-out orchestrator; respects max_concurrent cap; optional queue.Put side-channel.
  • feat(ptc/modules): modules/common/recovery_loop.ainl — bounded retry wrapper for ptc_runner calls based on ok: false envelope; max_attempts capped at 10.
  • feat(ptc/intelligence): intelligence/signature_enforcer.py — parse # signature: ... metadata, validate output shapes, run_with_signature_retry (max 3 attempts).
  • feat(ptc/intelligence): intelligence/trace_export_ptc_jsonl.py — export AINL trajectories in PTC-compatible JSONL; _strip_private_keys; beam_metrics and beam_telemetry fields.
  • feat(ptc/intelligence): intelligence/context_firewall_audit.py — audit .ainl source and trajectory files for _-prefixed key leakage; runnable via scripts/run_intelligence.py.
  • feat(ptc/intelligence): intelligence/ptc_to_langgraph_bridge.py — analyze .ainl source or trajectory for ptc_runner calls; emit LangGraph-compatible create_ptc_tool_node Python snippet.
  • feat(ptc/mcp): scripts/ainl_mcp_server.py — new tools: ainl_ptc_signature_check, ainl_trace_export, ainl_ptc_run, ainl_ptc_health_check, ainl_ptc_audit (combined signature + firewall report); wired into tooling/mcp_exposure_profiles.json.
  • feat(ptc/cli): cli/main.py — ainl run-hybrid-ptc subcommand: thin wrapper over ainl run with pre-configured defaults (mock mode, ptc_runner adapter, trace path); --no-mock and --trace-jsonl overrides.
  • feat(ptc/examples): examples/hybrid_order_processor.ainl — production hybrid example: parallel order batches, signature validation, recovery loop, _ context firewall, trace/LangGraph bridge comments.
  • feat(ptc/examples): examples/price_monitor.ainl — second production example: parallel price monitoring with ptc_parallel + recovery_loop + _ context firewall.
  • feat(ptc/docs): docs/assets/ptc_flow.mmd — Mermaid flow diagram: AINL graph → PTC modules → PTC Runner/BEAM → enriched envelope → trace export / LangGraph bridge → deployment.
  • feat(ptc/security): tooling/security_profiles.json — ptc_sandbox_plus named profile with max_subagent_budget, trace_depth, max_concurrent_calls.
  • docs(ptc): docs/adapters/PTC_RUNNER.md — full integration guide (setup, verbs, envelope, signature, context firewall, parallel, recovery, subprocess BEAM mode, LangGraph emission, MCP tools, Mermaid diagram, production examples).
  • docs(ptc): WHITEPAPERDRAFT.md §8.4 — PTC-Lisp hybrid integration positioning note.
  • docs(ptc): README.md, docs/README.md, docs/INTELLIGENCE_PROGRAMS.md, docs/EXAMPLE_SUPPORT_MATRIX.md, docs/operations/EXTERNAL_ORCHESTRATION_GUIDE.md — cross-references and quick-start callouts.

v1.2.8 (March 25, 2026) — OpenClaw ops, intelligence hydration, graph-runtime docs

  • docs(whitepaper): WHITEPAPERDRAFT.md (v1.2.8 positioning, §6.6 graph pitfalls, §10.5 intelligence, §13.5 token caps, §17.1 shipped, appendix OpenClaw file map); supporting updates: docs/WHAT_IS_AINL.md, docs/DOCS_INDEX.md, docs/overview/README.md, docs/POST_RELEASE_ROADMAP.md
  • packaging: pyproject.toml / ainl 1.2.8; RUNTIME_VERSION 1.2.8 in runtime/engine.py (mirrored tests/emits/server/runtime/engine.py); CITATION.cff aligned. Reinstall or pip install -U -e . so CLI/MCP/runner runtime_version matches and __pycache__ from older trees does not shadow updated modules.
  • feat(ops): rolling budget hydration for scripts/run_intelligence.py, tooling/intelligence_budget_hydrate.py, workspace env pin tooling/openclaw_workspace_env.example.sh, ops docs (docs/operations/*: profiles, token observability, workspace isolation, WASM/TTL/embedding notes).
  • docs(openclaw): docs/operations/OPENCLAW_AINL_GOLD_STANDARD.md — agent-discoverable install/upgrade checklist (profiles, caps, cron, host bootstrap, verification); indexed from tooling/bot_bootstrap.json, HOST_PACK_OPENCLAW.md, DOCS_INDEX.md.
  • docs(openclaw): docs/operations/OPENCLAW_HOST_AINL_1_2_8.md — host briefing for v1.2.8 (repo capabilities vs OpenClaw obligations: probe, rolling hydrate, profiles, bootstrap contract); openclaw_host_ainl_1_2_8 in tooling/bot_bootstrap.json.
  • fix(graph): intelligence + modules/common/generic_memory.ainl — graph-safe X (no raw {…} literals), memory.list optional prefix via null, metadata valid_at / tags; see docs/RUNTIME_COMPILER_CONTRACT.md § graph pitfalls, docs/AINL_SPEC.md.

v1.2.7 (March 24, 2026) — Hyperagent Research Pack (additive)

  • feat(cli): ainl inspect <file.ainl> [--strict] [--json] — dump full canonical IR JSON.
  • feat(runtime): ainl run --trace-jsonl PATH|- — structured JSONL execution tape (file or stdout).
  • feat(diagnostics): structured diagnostics expose llm_repair_hint for LLM-native repair loops.
  • feat(mcp): ainl_fitness_report and ainl_ir_diff tools for selection/mutation loops; ainl_fitness_report includes fitness_score with fitness_components/weights, plus adapter/operation/frame-key proxy metrics; ainl_ir_diff detects payload-level node data deltas (not only topology).
  • feat(schema): machine-readable schema seed ainl.schema.json.
  • docs: docs/EMBEDDING_RESEARCH_LOOPS.md, docs/operations/MCP_RESEARCH_CONTRACT.md, prompts/meta-agent/*.

v1.2.6 (March 24, 2026) — sandbox install hardening, wheel integrity gates, doctor command

  • fix(graph / intelligence): intelligence/token_aware_startup_context.lang and modules/common/generic_memory.ainl — avoid X {…} object literals (graph IR → unknown X fn: {); build filters/payloads with core.parse, obj/put, arr; merge list steps into one label (do not use J NextLabel as a jump); memory.list uses null for omitted record_id_prefix; memory_tags via X … (arr …); valid_at from R core iso. Documented in docs/AINL_SPEC.md, docs/RUNTIME_COMPILER_CONTRACT.md, docs/INTELLIGENCE_PROGRAMS.md, docs/adapters/MEMORY_CONTRACT.md § 3.4.
  • fix(lang/samples): intelligence programs, demos, and autonomous-ops examples — bind memory-contract frame variables with Set (memory_namespace, memory_kind, memory_record_id, …), not X, which parses the next token as fn (fixes unknown X fn / invalid core.isNull usage in scheduled OpenClaw flows). Documented in docs/AINL_SPEC.md, docs/INTELLIGENCE_PROGRAMS.md, docs/adapters/MEMORY_CONTRACT.md; module example comments in modules/common/access_aware_memory.ainl.
  • fix(packaging): include runtime and runtime.* in setuptools package discovery so wheel installs include runtime.compat and avoid wheel-only import regressions.
  • fix(skill-installer): skills/ainl/install.sh hardened for restricted environments (PEP 668 fallbacks), no eval, and idempotent PATH hint updates across .bashrc/.zshrc/.profile.
  • feat(cli): add ainl doctor for environment diagnostics (Python/import health, PATH checks, MCP config checks, install-mcp --dry-run checks).
  • feat(ci): add Linux/macOS install smoke matrix for Python 3.10–3.13 with normal, --user, and --break-system-packages install paths, plus pip check, ainl --help, and ainl-mcp --help.
  • feat(ci/release): add wheel-integrity CI and dedicated release gates (twine check, wheel import smoke for runtime.compat/adapters/cli.main, pip check, MCP dry-run installers).
  • feat(packaging/docs): add tested Python 3.13 MCP constraints (constraints/py313-mcp.txt) and document no-root install order (venv -> --user -> --break-system-packages) across install/release/skill docs.
  • feat(compiler/ir): add optional execution_requirements metadata (including avm_policy_fragment, isolation/capability/resource hints) for AVM/general sandbox handoff.
  • feat(runtime/sandbox): add unified optional sandbox shim (runtime/sandbox_shim.py) and light wiring in runner, MCP, and CLI runtime path with graceful fallback when no sandbox runtime is detected.
  • feat(cli): add ainl generate-sandbox-config <file.ainl> [--target avm|firecracker|gvisor|k8s|general].
  • feat(trajectory): add optional trajectory JSONL fields when sandbox shim is connected (avm_event_hash, sandbox_session_id, sandbox_provider, isolation_hash).
  • docs(integration): update integration/ops/getting-started docs and whitepaper references for AVM + general sandbox deployment posture.

v1.2.5 (March 23, 2026) — Hyperspace bridge + hybrid S, CI benchmark baselines, LangGraph emit

  • feat(compiler): S hybrid langgraph, S hybrid temporal, or both on one line — opt-in hybrid wrapper targets for minimal_emit / capability planning; stored as services.hybrid.emit (de-duped); strict mode rejects unknown targets
  • feat(tooling): legacy infer_artifact_capabilities (IRs without emit_capabilities) derives needs_langgraph / needs_temporal from services.hybrid.emit
  • feat(ci): benchmark-regression prefers committed tooling/benchmark_size_ci.json / tooling/benchmark_runtime_ci.json on the baseline SHA when present; falls back to full JSON reports; several jobs pinned to Python 3.10; make benchmark-ci prints resolved PYTHON
  • fix(emit/langgraph): emitted AinlHybridState uses plain dict fields so LangGraph’s get_type_hints path works on Python 3.10
  • test: hybrid e2e (StateGraph.invoke, Temporal ActivityEnvironment), S hybrid and emission-planner legacy IR coverage
  • docs: BENCHMARK.md (CI baseline rules), docs/AINL_SPEC.md / docs/language/grammar.md (S hybrid), docs/RELEASING.md (benchmark-ci commit step); expanded narrative docs/RELEASE_NOTES.md § v1.2.5; cross-links in README.md, docs/README.md, CONTRIBUTING.md, WHITEPAPERDRAFT.md, docs/POST_RELEASE_ROADMAP.md, docs/DOCS_INDEX.md, docs/WHAT_IS_AINL.md, WHAT_IS_AINL.md, docs/overview/README.md

Hyperspace bridge (trajectory, common modules, local adapters, emitter)

  • feat(runtime): optional per-step trajectory JSONL (--log-trajectory / AINL_LOG_TRAJECTORY) — docs/trajectory.md
  • feat(modules): modules/common/guard.ainl, session_budget.ainl, reflect.ainl — token/time ceilings, budget accounting, reflect gates; modules/common/README.md updated
  • feat(adapters): local vector_memory (keyword-scored JSON store) and tool_registry (JSON tool catalog) under adapters/; CLI --enable-adapter vector_memory|tool_registry; documented in docs/reference/ADAPTER_REGISTRY.md, docs/adapters/README.md
  • feat(emit): --emit hyperspace via scripts/validate_ainl.py / ainl-validate — compiler_v2.AICodeCompiler.emit_hyperspace_agent emits a single-file Python agent: base64 JSON IR blob (_IR_B64), repo-root discovery for runtime/ + adapters/, AdapterRegistry with core, vector_memory, tool_registry, RuntimeEngine (graph-preferred, step_fallback=True), run_ainl, __main__; AINL_LOG_TRAJECTORY → <source-stem>.trajectory.jsonl in cwd; optional hyperspace_sdk import with RuntimeWarning when absent (scaffold TODO for future SDK bridge)
  • feat(examples/teaching): examples/hyperspace_demo.ainl, examples/test_adapters_full.ainl — strict-valid dotted-verb demos; tooling/canonical_curriculum.json / scripts/build_canonical_training_pack.py outputs include these lessons
  • test(emit): hyperspace (and related) emitter coverage in tests/test_snapshot_emitters.py
  • docs: docs/emitters/README.md, root README.md; hub links in docs/README.md, docs/DOCS_INDEX.md; docs/runtime/README.md (trajectory); docs/examples/README.md (phase-2 / hyperspace examples); docs/language/AINL_CORE_AND_MODULES.md §8 (guard/budget/reflect); docs/INTELLIGENCE_PROGRAMS.md (infrastructure_watchdog.lang); docs/RUNTIME_COMPILER_CONTRACT.md (trajectory note); docs/RELEASE_NOTES.md § v1.2.5 — expanded Hyperspace emitter narrative
  • ops / examples: intelligence/infrastructure_watchdog.lang, scripts/morning_briefing.ainl + wrapper, small examples/test_*.ainl harnesses; .gitignore runtime_runner.log
  • docs (primer policy): docs/WHAT_IS_AINL.md is the canonical “What is AINL?” document (stakeholder narrative + capability snapshot); repository root WHAT_IS_AINL.md is a stub pointing to it. README.md, docs/README.md, docs/DOCS_INDEX.md, docs/overview/README.md, WHITEPAPERDRAFT.md (§5.4 includes note, new §6.5 trajectory / Hyperspace) updated accordingly.

Strict label dataflow and hybrid E2E (shipped with v1.2.5 line)

  • fix(strict): propagate_inter_label_entry_defs / merged entry defs in tooling/effect_analysis.py and compiler_v2._validate_graphs — variables live on all paths into a jumped-to label are recognized at that label (fewer false undefined-at-use errors on branch/cron monitor patterns)
  • test: tests/test_inter_label_dataflow.py; Temporal execute_workflow path in tests/test_hybrid_emit_integration.py (WorkflowEnvironment, Worker, ThreadPoolExecutor activity executor)
  • tooling: tooling/artifact_profiles.json strict-valid refresh (examples/cron/monitor_and_alert.ainl, corpus/example_monitor_alert/program.ainl, …); curriculum order bump + regenerated tooling/canonical_training_pack.json / tooling/training_packs/*; conformance syrupy snapshots under tests/snapshots/conformance/

v1.2.4 (March 21, 2026)

Addendum (2026-03-22).

  • fix(apollo-x-bot): gateway_server.py — _classify_wants_envelope only when messages is a non-empty list (avoids envelope_missing_messages on classify_response=raw without messages; legacy tweet+prompt path); clearer bind error on EADDRINUSE; swallow BrokenPipeError / connection resets when the client disconnects early (timeouts).

  • fix(cli): ainl run --http-timeout-s help text — note LLM / bridge latency; default remains 5 (callers with slow routes must raise it).

  • fix(apollo-x-bot): openclaw-poll.sh / run-with-gateway.sh — --http-timeout-s 120, overridable via AINL_HTTP_TIMEOUT_S.

  • fix(graph): apollo-x-bot/ainl-x-promoter.ainl — promoter.process_tweet payload: stringify tweet directly for the bridge body; skip branch uses boolean true for JSON ok (strict-safe).

  • test: tests/test_apollo_x_gateway.py — regression for classify raw without messages; expanded gateway coverage as applicable.

  • fix(compiler_v2): strict reachability — treat If as a valid label terminator (skip “exactly one J” false positive when If is last); include If when wiring Loop/While fall-through edges to following labels.

  • fix(demo): demo/infrastructure_watchdog.lang — restart services when down; verify after restart before alerting.

  • docs: docs/reference/ADAPTER_REGISTRY.md §2.4.3 (bridge client timeout); docs/integrations/EXTERNAL_EXECUTOR_BRIDGE.md §7 (timeouts + llm.classify envelope rule); docs/OPENCLAW_INTEGRATION.md (Apollo promoter + openclaw-poll.sh pointer); apollo-x-bot/OPENCLAW_DEPLOY.md (AINL_HTTP_TIMEOUT_S / --http-timeout-s).

  • packaging: pyproject.toml / ainl 1.2.4; RUNTIME_VERSION (runtime/engine.py, mirrored tests/emits/server/runtime/engine.py) 1.2.4; language server serverInfo.version and runner service FastAPI app.version follow RUNTIME_VERSION (runner: scripts/runtime_runner_service.py; LSP: langserver.py)

  • feat(modules): modules/common/access_aware_memory.ainl — opt-in LACCESS_READ, LACCESS_WRITE, LACCESS_LIST, LACCESS_LIST_SAFE (graph-safe list touches via While + index); header warnings and usage notes for graph vs ForEach

  • fix(runtime): resolve bare label targets against include alias from call stack (_resolve_label_key in runtime/engine.py) so nested If / Loop / Call / While reach alias/label keys after merge

  • docs: modules/common/README.md; refresh root README.md, WHAT_IS_AINL.md, docs/WHAT_IS_AINL.md, WHITEPAPERDRAFT.md, docs/RELEASE_NOTES.md, docs/POST_RELEASE_ROADMAP.md, docs/RUNTIME_COMPILER_CONTRACT.md, docs/README.md, docs/adapters/README.md, docs/DOCS_INDEX.md, docs/operations/EXTERNAL_ORCHESTRATION_GUIDE.md (sample runtime_version in JSON)

  • test: tests/test_demo_enforcer.py (demo compile + access-aware smoke; graph-only bare-label regression)

  • fix(programs): S is parsed as service mode path (three slots); patterns like S core memory cron "0 * * * *" wrongly set path to cron and drop the schedule. Restored S core cron "<expr>" (and equivalent) in intelligence/token_aware_startup_context.lang, intelligence/session_continuity_enhanced.lang, intelligence/proactive_session_summarizer.lang, demo/session_continuity.lang, and examples/autonomous_ops/memory_prune.lang.

  • fix(programs): examples/autonomous_ops/meta_monitor.lang — replace invalid S cron + Cr with S cache cron "*/15 * * * *" (aligned with demo/meta_monitor.lang).

  • tooling: scripts/validate_s_cron_schedules.py and console script ainl-validate-s-cron (pyproject.toml) — fail CI/local runs on malformed S+cron lines; tests/test_s_cron_schedule_lines.py keeps the rule green.

  • docs: docs/CRON_ORCHESTRATION.md — new § S line shape (cron schedules) and § Security: queues, notifications, and secrets; docs/RUNTIME_COMPILER_CONTRACT.md, docs/DOCS_INDEX.md; web mirror ainativelangweb/content/docs/RUNTIME_COMPILER_CONTRACT.mdx.

  • docs(intelligence): intelligence/token_aware_startup_context2.lang — explicit legacy header; prefer token_aware_startup_context.lang for production (budget gate + structured memory + access-aware reads).

  • bench: regenerate BENCHMARK.md, tooling/benchmark_size.json, tooling/benchmark_runtime_results.json, and CI twins tooling/benchmark_size_ci.json / tooling/benchmark_runtime_ci.json via make benchmark / make benchmark-ci; refresh README.md and docs/benchmarks.md headline table (e.g. 19 strict-valid, public_mixed path counts + viable ratios; make benchmark uses --mode wide: full_multitarget_core + full_multitarget + minimal_emit).

  • interop: optional [interop] extra in pyproject.toml (langgraph, temporalio, aiohttp); [benchmark] also lists temporalio for hybrid smoke tests — see docs/PACKAGING_AND_INTEROP.md.

v1.2.3 (March 20, 2026)

  • bench(docs): major size benchmark documentation refresh (Mar 2026): default tiktoken cl100k_base in BENCHMARK.md tables; viable subset vs legacy-inclusive dual reporting; minimal_emit python_api fallback stub and prisma / react_ts compaction called out in transparency notes; JSON schema 3.5+; hub docs/benchmarks.md highlights table; README / WHITEPAPERDRAFT.md §12 / cross-links updated. Representative headline: public_mixed minimal_emit ~1.02× on viable rows (46/59); legacy-inclusive minimal ~0.24× shown separately.
  • docs(ops): add docs/operations/UNIFIED_MONITORING_GUIDE.md (unified AINL + OpenClaw monitoring); expand docs/openclaw/BRIDGE_TOKEN_BUDGET_ALERT.md; cross-link hub from docs/README.md, docs/DOCS_INDEX.md, docs/operations/README.md, docs/CRON_ORCHESTRATION.md, docs/AINL_CANONICAL_CORE.md, docs/ainl_openclaw_unified_integration.md, docs/adapters/*, docs/BOT_ONBOARDING.md, docs/OPENCLAW_IMPLEMENTATION_PREFLIGHT.md, docs/operations/AUTONOMOUS_OPS_PLAYBOOK.md, docs/operations/AUTONOMOUS_OPS_MONITORS.md, docs/advanced/README.md, docs/adapters/MEMORY_CONTRACT.md; standardize daily memory path ~/.openclaw/workspace/memory/YYYY-MM-DD.md; refresh openclaw/bridge/README.md monitoring tables and env reference
  • feat(modules): add reusable memory helper include modules for production monitor flows:
    • modules/common/token_cost_memory.ainl for deterministic workflow namespace writes/lists
    • modules/common/ops_memory.ainl for deterministic ops namespace writes/lists
  • refactor(programs): migrate monitor-heavy demo/ and examples/autonomous_ops/ workflows from repeated inline memory logic to shared include calls while preserving payloads, record kinds, tags, TTLs, and control-flow behavior
  • feat(programs): normalize deterministic filters across migrated history reads (updated_after, tags_any, source, limit) and standardize metadata envelopes on writes (source, confidence, tags, valid_at)
  • docs(ops): refresh autonomous ops documentation to reflect include-helper architecture and current monitor implementation posture
  • feat(strict): allowlist memory.PUT / GET / APPEND / LIST / DELETE / PRUNE in tooling/effect_analysis.py and treat R memory <verb> … steps as memory.<VERB> for strict adapter checks
  • test(conformance): add memory_continuity_runtime snapshot category and tests/data/conformance/session_budget_memory_trace.ainl; extend tokenizer round-trip with demo/session_budget_enforcer.lang and IR canonicalization for the trace golden
  • docs(examples): add examples/timeout_memory_prune_demo.ainl, committed docs/assets/timeout_memory_prune_flow.png, README Memory & State pointer to demo/session_budget_enforcer.lang, and visualizer PNG example command

v1.2.2 (March 20, 2026)

  • feat(memory): implement Memory Contract v1.1 additive runtime support in runtime/adapters/memory.py:
    • optional deterministic metadata envelope fields (source, confidence, tags, valid_at)
    • deterministic memory.list bounded filters (tags_any, tags_all, created/updated windows, source, valid_at windows, limit, offset)
    • namespace retention hooks (default_ttl_by_namespace, prune_strategy_by_namespace)
    • portable operational counters in adapter responses (stats.operations, stats.reads, stats.writes, stats.pruned)
  • feat(capabilities): add capability-advertised memory profile in tooling/capabilities.json (memory_profile: "v1.1-deterministic-metadata") with schema support in tooling/capabilities.schema.json
  • test(memory): extend tests/test_memory_adapter.py for metadata round-trip, deterministic list filters/pagination, retention hooks, and operational counters
  • docs(memory): align docs/adapters/MEMORY_CONTRACT.md and docs/reference/CAPABILITY_REGISTRY.md with shipped v1.1 additive behavior

v1.2.1 (March 20, 2026)

  • feat(visualizer): add image export for ainl visualize via Playwright (--png, --svg, --width, --height, and extension auto-detect for .png/.jpg/.jpeg/.svg)
  • test(visualizer): add tests/test_visualizer.py smoke coverage for PNG/SVG generation
  • docs(examples): add examples/timeout_demo.ainl and expand README/docs guidance for starter timeout module and visualizer image export usage
  • fix(readme): remove duplicate badge block and switch release badge to latest tag endpoint

v1.2.0 (March 20, 2026)

  • feat: subgraph includes & modules (compile-time composition, alias prefixing, ENTRY/EXIT contract)
  • feat: graph visualizer CLI (ainl visualize → Mermaid with clusters, synthetic call edges)
  • feat: full conformance matrix runner (make conformance) with snapshot categories for tokenizer, IR, strict validation, runtime parity, and emitter stability
  • fix: preserve __literal_fields through include merge + lowering
  • docs: updated README/docs hubs for conformance command, CI gate, and generated conformance artifacts

Packaging (v1.2.0)

  • pyproject.toml / ainl: version 1.2.0.
  • RUNTIME_VERSION (runtime/engine.py, mirrored tests/emits/server/runtime/engine.py): 1.2.0 (runner/MCP payloads, /capabilities, ainl run --json).
  • Language server initialize serverInfo.version and runner service OpenAPI app version: 1.2.0.

Highlights

  • Includes (compile-time composition): top-level include path [as alias] merges submodule labels into the parent IR as alias/LABEL (e.g. retry/ENTRY, retry/EXIT_OK). Strict mode enforces ENTRY / EXIT_* shapes for included graphs. Starter modules: modules/common/retry.ainl, modules/common/timeout.ainl. Tests: tests/test_includes.py.
  • Graph visualizer CLI: ainl visualize / ainl-visualize / scripts/visualize_ainl.py — strict compile → Mermaid graph TD from ir["labels"]; subgraph clusters per include alias; synthetic Call → callee entry edges with a %% comment; flags --no-clusters, --labels-only, -o -. Fixture: examples/bad_include.ainl.
  • Structured diagnostics: native Diagnostic rows, --diagnostics-format (auto | plain | rich | json), merge dedup (native-first), optional rich stderr with pip install -e ".[dev]" — used by ainl-validate and ainl visualize (and the language server path).
  • Strict literals / dataflow: quote string payloads where strict dataflow expects literals (e.g. J "ok"); aligned with docs/RUNTIME_COMPILER_CONTRACT.md and conformance tests.

Documentation

  • Root README.md (quick-start write → validate → visualize → run, includes, visualize flags, docs/WHAT_IS_AINL.md link), WHAT_IS_AINL.md, docs/WHAT_IS_AINL.md, WHITEPAPERDRAFT.md (v1.2.0, §5.4–5.5), docs/POST_RELEASE_ROADMAP.md, docs/CONFORMANCE.md, SEMANTICS.md, docs/architecture/GRAPH_INTROSPECTION.md, docs/INSTALL.md, docs/DOCS_INDEX.md, docs/overview/README.md, docs/README.md.

Tooling — graph visualizer CLI (Mermaid) (2026-03-20)

Features

  • scripts/visualize_ainl.py — strict compile → Mermaid graph TD from ir["labels"]; subgraph clusters by include alias; synthetic Call → entry edges with %% comment; --no-clusters, --labels-only; structured diagnostics on failure (reuse validate_ainl formatters).
  • ainl-visualize console script and ainl visualize subcommand (cli/main.py, pyproject.toml).
  • examples/bad_include.ainl — fixture for visualize/strict error demos.

Documentation

  • Root README.md (Visualize your workflow, mermaid.live walkthrough), .gitignore (hello.mmd).
  • docs/architecture/GRAPH_INTROSPECTION.md §7 (Mermaid), §8 (DOT; former §6), fixed duplicate §5 → §6 record/replay.
  • docs/INSTALL.md, docs/DOCS_INDEX.md, docs/README.md, docs/architecture/README.md, docs/reference/GRAPH_SCHEMA.md.

Compiler — diagnostics merge dedup + --diagnostics-format (2026-03-09)

Behavior

  • compiler_v2: After merging native and string-derived diagnostics, deduplicate by (kind, message body) with Line N: stripped; native rows first so spans, label_id, and suggestions win over parsed duplicates.
  • ainl-validate: --diagnostics-format=auto|plain|json|rich (default auto). --json-diagnostics remains an alias for json. --no-color forces plain output instead of rich.

Documentation

  • README.md, docs/INSTALL.md.

Compiler — structured diagnostics + ainl-validate CLI (2026-03-09)

Features

  • compiler_diagnostics.py — Diagnostic, CompilerContext, CompilationDiagnosticError; native diagnostics merged with legacy errors strings.
  • Strict-mode native diagnostics in compiler_v2.py for: arity (min_slots), unknown module prefix, duplicate Lxx: (strict-only legacy error; non-strict still merges bodies), undeclared endpoint label, undeclared targeted label (control-flow refs). Spans, suggested_fix, related_span, and label close-match hints where applicable.
  • ainl-validate / scripts/validate_ainl.py — --strict wires structured diagnostics by default; stderr shows a numbered report (3-line source context, carets for spans). --json-diagnostics prints diagnostics JSON only to stdout on failure. rich (optional dev extra) enables styled output; plain ANSI or no-color when rich is missing or --no-color / non-TTY.
  • Tests: tests/test_diagnostics.py (structured round-trip, phase-2 sites, formatter smoke).

Documentation

  • README.md, docs/INSTALL.md, docs/architecture/GRAPH_INTROSPECTION.md, docs/CONFORMANCE.md, docs/POST_RELEASE_ROADMAP.md, pyproject.toml (rich in [project.optional-dependencies] dev).

Documentation — agent reports & intelligence hub (2026-03-19)

Documentation

  • agent_reports/README.md — index of OpenClaw field reports (distinct from CONSULTANT_REPORTS.md / AI_CONSULTANT_REPORT_*.md).
  • docs/INTELLIGENCE_PROGRAMS.md — map of intelligence/*.lang, host responsibilities, and scripts/run_intelligence.py.
  • scripts/run_intelligence.py — dev runner for context / summarizer / consolidation / continuity / all (OpenClaw monitor registry).
  • CONSULTANT_REPORTS.md, docs/DOCS_INDEX.md, docs/AI_AGENT_CONTINUITY.md, AI_AGENT_QUICKSTART_OPENCLAW.md, README.md — cross-links to the above.
  • Field report: agent_reports/ainl-king-openclaw-2026-03-19.md (Day 2 — AINL King; merged via PR #2).

1.1.1 — Runtime correctness fixes and adapter extensions (2026-03-09)

Bug fixes

  • Compiler: X-step S-expression paren stripping — X dst (core.add 3 4) and similar paren-form X expressions now parse correctly. The tokenizer does not treat ( and ) as delimiters, so the leading ( was attaching to the function name and the entire core.* engine branch was unreachable dead code. Fixed in both X-step parser paths in compiler_v2.py. Snapshot updated for examples/openclaw/daily_digest.lang which uses (core.div ...) and (core.gt ...).

  • Runtime: J variable resolution — J var was returning the raw token string "var" instead of resolving var from the execution frame. Fixed with a one-line _resolve() call in runtime/engine.py. This was a silent correctness bug since the original release; every program using J for its return value was affected. Snapshots updated.

New adapter verbs (OpenClaw extension level)

  • web.search — new WebAdapter in adapters/openclaw_integration.py backed by OpenRouter/Perplexity. Returns search results for use in intelligence digest and monitoring workflows. Requires OPENROUTER_API_KEY.

  • tiktok.recent — new verb on the TikTokAdapter for retrieving recent posts/metrics with a configurable limit.

  • core.idiv — integer division builtin added to CoreBuiltinAdapter in runtime/adapters/builtins.py. Available as both R core.idiv and X dst (core.idiv a b) in X expressions.

  • memory ops namespace — ops added to the valid namespace whitelist in the memory adapter for operational metrics and monitor state storage.

Dependency

  • requests>=2.28.0 declared under [project.optional-dependencies] openclaw in pyproject.toml. The WebAdapter uses it; previously relied on silently as a transitive dep.

Documentation

  • docs/INSTALL.md: recommended Python 3.10 layout using .venv-py310 and PYTHON_BIN=python3.10 bootstrap; agents/CI parity note.

  • scripts/bootstrap.sh: PYTHON_BIN / VENV_DIR env vars + 3.10 minimum check.

  • scripts/precommit_docs_contract.sh + .pre-commit-config.yaml: docs contract hook resolves Python via .venv-py310, then .venv, then python3 (no bare python on PATH required).

  • docs/AINL_SPEC.md: X, Loop, While, ForEach added to grammar and slot rules tables; J description clarified to say "resolve from frame".

  • docs/ainl_runtime_spec.md: X fn list updated (idiv, concat, join, ite/if, core.* prefix form, paren syntax); J semantics note added.

  • docs/reference/ADAPTER_REGISTRY.md: web adapter added; tiktok.recent and core.idiv/IDIV documented; memory ops namespace noted.

  • docs/adapters/MEMORY_CONTRACT.md: ops namespace added to v1 whitelist.

1.1.0 — First Public Release Candidate (2026-03-09)

Baseline

  • Official Python minimum raised to 3.10+; metadata, docs, bootstrap, and CI (3.10 + 3.11) aligned.
  • Core test profile fully green (403 tests, 0 failures).

MCP Server (v1)

  • Added scripts/ainl_mcp_server.py — a thin, stdio-only MCP server exposing ainl_validate, ainl_compile, ainl_capabilities, ainl_security_report, and ainl_run as MCP tools, plus ainl://adapter-manifest and ainl://security-profiles as MCP resources.
  • Safe-default posture: core-only adapters, conservative limits, hardcoded local_minimal-style policy.
  • Quickstart and end-to-end example in docs/operations/EXTERNAL_ORCHESTRATION_GUIDE.md section 9.

Security & Operator Surfaces

  • Adapter privilege-tier metadata in tooling/adapter_manifest.json.
  • Policy validator supports forbidden_privilege_tiers.
  • Named security profiles in tooling/security_profiles.json.
  • Security/privilege introspection via tooling/security_report.py.
  • Sandbox, orchestration, and threat-model docs shipped.

Documentation

  • docs/ reorganized into intent-based sections (overview, fundamentals, getting_started, architecture, runtime, language, adapters, emitters, examples, case_studies, competitive, operations, advanced, reference) with section READMEs, compatibility stubs, and a root navigation hub.
  • README: added "Choose Your Path" mode chooser (CLI / runner / MCP) with a single canonical example shown three ways.
  • Getting-started guide enriched with integration paths and core-first guidance.
  • Release notes finalized with current milestone summary.

Polish

  • Runner service migrated from deprecated @app.on_event to FastAPI lifespan handlers.
  • Memory tooling migrated from datetime.utcnow() to timezone-aware UTC helpers.
  • Repo hygiene: scratch files removed, .gitignore tightened.

Full details: docs/RELEASE_NOTES.md.

1.0.15-memory-v1-and-interoperability (2026-03-09)

OpenClaw Autonomous Ops Programs and Docs

  • Added new autonomous ops programs (OpenClaw-created): memory_prune, meta_monitor, session_continuity, token_budget_tracker, and monitor_system (reference shape). Demo and examples/autonomous_ops/ copies added; existing monitors (canary_sampler, infrastructure_watchdog, lead_quality_audit, tiktok_sla_monitor, token_cost_tracker) updated.
  • New docs: docs/operations/AUTONOMOUS_OPS_MONITORS.md (monitor index and schedules), docs/operations/STANDARDIZED_HEALTH_ENVELOPE.md. OpenClaw implementation notes: openclaw/MEMORY_PRUNE_IMPLEMENTATION.md, openclaw/META_MONITOR_IMPLEMENTATION.md, openclaw/SESSION_CONTINUITY_IMPLEMENTATION.md, openclaw/TOKEN_BUDGET_TRACKER_IMPLEMENTATION.md, plus update docs for existing monitors.
  • Updated docs/DOCS_INDEX.md (links to autonomous ops monitors and standardized health envelope), docs/EXAMPLE_SUPPORT_MATRIX.md and examples/autonomous_ops/README.md (new programs table), tooling/artifact_profiles.json (new example paths). Config externalization and related OpenClaw notes in openclaw/CONFIG_EXTERNALIZATION.md, openclaw/CANARY_SAMPLER_UPDATE.md, etc.

Extension-Level Memory Adapter and Contract (v1)

  • Added an extension-level memory adapter backed by SQLite with explicit record identity (namespace, record_kind, record_id) and three core verbs:
    • memory.put(namespace, record_kind, record_id, payload, ttl_seconds?)
    • memory.get(namespace, record_kind, record_id)
    • memory.append(namespace, record_kind, record_id, entry, ttl_seconds?)
  • Documented the v1 contract in docs/adapters/MEMORY_CONTRACT.md, including:
    • namespace whitelist and recommended record kinds,
    • advisory TTL semantics,
    • validation expectations and backend schema.
  • Marked the adapter as extension_openclaw / non-canonical in tooling/adapter_manifest.json and docs/reference/ADAPTER_REGISTRY.md.

Memory Validator/Linter and JSON Bridges

  • Added an extension-only validator/linter for memory envelopes:
    • tooling/memory_validator.py
    • scripts/validate_memory_records.py
  • Implemented JSON/JSONL export/import tooling on top of the SQLite store:
    • tooling/memory_bridge.py
    • scripts/export_memory_records.py
    • scripts/import_memory_records.py
  • These tools:
    • treat the memory_records table as source of truth,
    • export canonical envelopes including provenance and flags,
    • import validated envelopes back into the store while preserving provenance and flags inside payload._provenance / payload._flags.

Markdown Bridges and Legacy Migration

  • Added a one-way, human-facing markdown export for daily logs:
    • tooling/memory_markdown_bridge.py
    • scripts/export_memory_daily_log_markdown.py
    • maps daily_log.note records to memory/daily_log/YYYY/YYYY-MM-DD.md.
  • Added curated, frontmatter-based markdown import for long-term kinds:
    • tooling/memory_markdown_import.py
    • scripts/import_memory_markdown.py
    • supports long_term.project_fact and long_term.user_preference only.
  • Added a narrow legacy migration helper to bootstrap from existing note habits:
    • tooling/memory_migrate.py
    • scripts/migrate_memory_legacy.py
    • migrates MEMORY.md sections to long_term.project_fact and memory/YYYY-MM-DD.md files to daily_log.note.

Discovery/Enumeration: memory.list

  • Extended the memory adapter with a fourth verb:
    • memory.list(namespace, record_kind?, record_id_prefix?, updated_since?)
  • Provides structured enumeration of records without returning payloads:
    • filters by namespace (required), optional kind, optional record_id prefix, and optional updated_since (ISO timestamp on updated_at),
    • returns deterministic, lightweight summaries (record_kind, record_id, created_at, updated_at, ttl_seconds).
  • Documented this as discovery-only, not a general query/search surface, and updated tests and docs to keep the adapter manifest, registry, and contract aligned.

Lifecycle Cleanup: memory.delete

  • Added an exact-key lifecycle verb:
    • memory.delete(namespace, record_kind, record_id)
  • Provides explicit deletion for a single record identified by its canonical key, returning {ok, deleted} and leaving neighboring records untouched.
  • Kept this strictly non-bulk and non-query: no predicates, no wildcard/batch deletion, and no TTL/prune semantics in this pass.

TTL Pruning: memory.prune

  • Added an operator-oriented TTL prune helper:
    • memory.prune(namespace?)
  • Removes only records whose ttl_seconds is set and whose created_at + ttl_seconds lies strictly in the past, optionally scoped to a single namespace.
  • Does not provide predicate-based/bulk deletion or broader retention policy logic; it is a small explicit cleanup tool on top of the existing TTL semantics.

1.0.14-advanced-coordination-governance (2026-03-09)

Safe Use, Threat Model, and Advanced Framing

  • Added docs/advanced/SAFE_USE_AND_THREAT_MODEL.md to describe:
    • what the local coordination substrate assumes (local-first, file-backed, sandboxed roots, external orchestrator),
    • what it does not provide (no built-in auth, encryption, policy, or multi-tenant isolation),
    • the threat model and trust assumptions for coordination usage,
    • and a clear split between supported safe-default use and advanced/unsupported-by-default patterns.
  • Updated README.md and docs/DOCS_INDEX.md to clearly separate:
    • core / safe-default surfaces (compiler/runtime/IR, canonical examples),
    • advanced / operator-only / experimental surfaces (OpenClaw extensions, coordination docs, and examples).

Coordination Validator and Mailbox Linter (extension-only)

  • Added an extension-only coordination schema validator and mailbox linter:
    • tooling/coordination_validator.py
    • scripts/validate_coordination_mailbox.py
  • The validator checks AgentTaskRequest and AgentTaskResult envelopes in JSON/JSONL files against the minimal upstream schema documented in docs/advanced/AGENT_COORDINATION_CONTRACT.md and reports malformed or drifted envelopes.
  • This tooling is optional, additive, and non-semantic: it does not change compiler/runtime behavior and is intended as governance and compatibility support for advanced coordination users.

Advanced Coordination Labeling and Release Checklist

  • Tightened framing for advanced coordination/OpenClaw examples:
    • added "advanced / operator-only / advisory-only" notes to key examples under examples/openclaw/,
    • updated docs/EXAMPLE_SUPPORT_MATRIX.md to describe the OpenClaw family as advanced, extension-only, and not a safe default for unsupervised agents.
  • Marked the agent adapter as advanced, opt-in in docs/reference/ADAPTER_REGISTRY.md and reiterated its extension-only, noncanonical status.
  • Extended docs/RELEASE_NOTES_DRAFT.md with a short "Advanced coordination" section clarifying that:
    • coordination is experimental/operator-only,
    • coordination is local/file-backed and advisory,
    • upstream does not claim it is a secure multi-agent or orchestration platform by default.
  • Updated docs/GITHUB_RELEASE_CHECKLIST.md with coordination-specific items:
    • protocol surface tests,
    • baseline verification,
    • mailbox validator run on baseline artifacts,
    • and a check that core vs advanced framing remains intact in README.md and docs/DOCS_INDEX.md.

1.0.13-coordination-baseline-and-monitor-advisory (2026-03-09)

Coordination Baseline (docs-only)

  • Added a human-visible coordination baseline section to docs/advanced/AGENT_COORDINATION_CONTRACT.md:
    • records the current baseline id: coordination-baseline-2026-03-09-token-and-monitor-advisory,
    • lists the required coordination artifacts (docs, examples, and sample result/snapshot JSON files) that both Cursor-side and OpenClaw-side environments should have on disk before controlled validation.
  • Clarified usage guidance:
    • missing or renamed coordination artifacts should be treated as repo-state drift, not as a protocol change,
    • protocol-shaping artifacts should not be silently recreated without updating the documented baseline in a follow-up change.

Monitor-Status Advisory Review Workflow

  • Added extension/OpenClaw examples for a bounded monitor-status advisory loop built on top of the existing agent coordination substrate:
    • examples/openclaw/monitor_status_advice_request.lang — assembles a small snapshot of monitor IDs, services, statuses, and recent last_ok / last_error timestamps, wraps it in an AgentTaskRequest, and enqueues it via agent.send_task (returning the task_id).
    • examples/openclaw/monitor_status_advice_read.lang — reads a single advisory AgentTaskResult for a known monitor-status task id via agent.read_result.
  • Added examples/openclaw/monitor_status_example_snapshot.json as a small, non-semantic JSON snapshot illustrating the payload shape used in the request example.
  • Updated docs/advanced/AGENT_COORDINATION_CONTRACT.md and docs/EXAMPLE_SUPPORT_MATRIX.md to describe and classify the new monitor-status advisory workflow as extension/OpenClaw, noncanonical, local-first, file-backed, sandboxed, and advisory-only.

1.0.12-local-agent-coordination-substrate (2026-03-09)

Local Agent Coordination Substrate (extension/OpenClaw)

  • Added a minimal local mailbox-style coordination adapter agent with a narrow shared protocol surface:
    • agent.send_task(envelope) -> task_id (string)
    • agent.read_result(task_id) -> AgentTaskResult (dict)
  • Implemented the adapter as a sandboxed, file-backed extension under AINL_AGENT_ROOT:
    • tasks are appended to tasks/openclaw_agent_tasks.jsonl,
    • results are read from results/<task_id>.json,
    • all paths are checked via a sandbox root helper, and attempts to escape the sandbox or use filesystem root are rejected.
  • Enforced that AgentTaskResult files must parse as JSON objects; invalid or non-object JSON now raise AdapterError.
  • Locked the shared protocol surface in tests (tests/test_agent_protocol_surface.py) so only send_task and read_result remain part of the coordination API.

Agent Coordination Contract and Examples

  • Added docs/advanced/AGENT_COORDINATION_CONTRACT.md as the canonical design/spec for:
    • AgentManifest
    • AgentTaskRequest
    • AgentTaskResult and the local mailbox protocol under AINL_AGENT_ROOT.
  • Documented the shared protocol boundary (Cursor ↔ OpenClaw) and explicitly excluded additional verbs such as read_task and list_agents from the shared surface.
  • Added small extension/OpenClaw examples:
    • examples/openclaw/agent_send_task.lang — build and enqueue a minimal AgentTaskRequest and return the task_id.
    • examples/openclaw/agent_read_result.lang — read a single AgentTaskResult identified by task_id.
    • examples/openclaw/token_cost_advice_request.lang — enqueue a bounded token-cost advisory request for a Cursor-side agent.
    • examples/openclaw/token_cost_advice_read.lang — read the advisory AgentTaskResult for a known token-cost task id.
  • Updated docs/reference/ADAPTER_REGISTRY.md and docs/EXAMPLE_SUPPORT_MATRIX.md to classify the agent adapter and the new examples as extension/OpenClaw, noncanonical surfaces.

Security Hardening

  • Hardened the agent adapter sandbox:
    • rejected AINL_AGENT_ROOT="/" (filesystem root) as an invalid sandbox,
    • rejected task_id values containing path separators or .. in agent.read_result,
    • ensured all file reads/writes go through a safe-path helper.
  • Added tests in tests/test_agent_send_task.py to assert:
    • sandbox escapes are rejected,
    • filesystem-root sandboxing is rejected,
    • invalid and non-object JSON results are rejected.

1.0.11-extras-metrics-and-graph-dot (2026-03-09)

Observability Helper: extras.metrics

  • Extended the OpenClaw extras adapter with a new metrics verb that reads precomputed JSON summaries (e.g. from scripts/summarize_runs.py) and exposes a small, stable metrics envelope:
    • run_count, ok_count, error_count
    • ok_ratio (when run_count > 0)
    • runtime_versions
    • result_kinds, trace_op_counts, label_counts
    • timestamps_present
  • Implemented extras.metrics as a sandboxed, read-only extension/OpenClaw helper:
    • summaries are resolved relative to AINL_SUMMARY_ROOT (default: /tmp/ainl_summaries),
    • attempts to escape the sandbox root raise AdapterError.
  • Kept extras.metrics clearly non-canonical and outside the strict core: it is classified as extension_openclaw, strict_contract: false, and documented in docs/reference/ADAPTER_REGISTRY.md.
  • Added tests/test_extras_metrics.py to lock in the envelope shape and error behavior for missing files, invalid JSON, and non-object summaries.

Graph DOT Export Helper

  • Added scripts/render_graph.py as a tiny helper to render compiled IR/graphs to DOT for visualization:
    • supports compiling .ainl / .lang files with AICodeCompiler or reading existing IR JSON,
    • groups IR nodes per label in DOT subgraphs,
    • labels nodes by op and adapter prefix (e.g. R (http)), and edges by control-flow port (next, then, else, etc.).
  • Documented the helper and example usage in docs/architecture/GRAPH_INTROSPECTION.md without changing any compiler or runtime semantics.
  • Added tests/test_render_graph.py to assert basic DOT structure and labeling.

Autonomous Ops Example Pack Expansion

  • Expanded examples/autonomous_ops/ with additional extension/OpenClaw monitors:
    • infrastructure_watchdog.lang
    • tiktok_sla_monitor.lang
    • lead_quality_audit.lang
    • token_cost_tracker.lang
    • canary_sampler.lang
  • Updated examples/autonomous_ops/README.md, docs/EXAMPLE_SUPPORT_MATRIX.md, and tooling/artifact_profiles.json to classify these as non-strict, extension/OpenClaw operational examples (not canonical core).

1.0.10-meta-monitoring-toolkits (2026-03-09)

Run Summary Helper

  • Added scripts/summarize_runs.py as a small meta-monitoring helper that aggregates one or more RuntimeEngine.run(..., trace=True) JSON payloads into a concise JSON summary:
    • run_count / ok_count / error_count
    • runtime_versions
    • result_kinds (Python type names of result)
    • trace_op_counts (per-op counts from traces)
    • label_counts (per-label trace event counts)
    • timestamps_present (currently false, since payloads do not include wall-clock timestamps)
  • Kept the tool strictly read-only over existing payloads; it does not change any language, compiler, or runtime semantics.
  • Added tests/test_summarize_runs.py to lock in the summary schema and basic aggregation behavior for single/multiple runs and list-of-payloads JSON input.
  • Extended docs/architecture/COMPILE_ONCE_RUN_MANY.md, README.md, and docs/DOCS_INDEX.md with small notes on how to use the new run-summary helper.

1.0.9-autonomous-ops-and-http-envelopes (2026-03-09)

Autonomous Ops Docs and Examples

  • Added docs/operations/AUTONOMOUS_OPS_PLAYBOOK.md as the central, truthful playbook for autonomous-agent use of AINL (compile-once/run-many, cooldown patterns, remediation flows, and current meta-monitoring limits).
  • Added examples/autonomous_ops/ with a small extension/OpenClaw autonomous-ops pack:
    • status_snapshot_to_queue.lang
    • backup_freshness_to_queue.lang
    • pipeline_readiness_snapshot.lang All are explicitly classified as extension_openclaw, non-canonical, non-strict-only snapshot emitters.
  • Updated examples/README.md, tooling/artifact_profiles.json, tooling/support_matrix.json, and docs/EXAMPLE_SUPPORT_MATRIX.md to classify the autonomous-ops pack as compatible OpenClaw extension examples.

Graph/IR Introspection and Operator UX

  • Added docs/architecture/GRAPH_INTROSPECTION.md to document IR/graph emission, graph API helpers, normalization, and diffs.
  • Added scripts/inspect_ainl.py as a tiny program-summary helper that prints graph_semantic_checksum, label/node counts, adapters, endpoints, and diagnostics without requiring users to read full IR JSON.
  • Updated README.md and docs/DOCS_INDEX.md to link to the new graph introspection and inspect helper surfaces.

Compile-Once / Run-Many Proof Pack

  • Added docs/architecture/COMPILE_ONCE_RUN_MANY.md with a minimal, reproducible recipe to:
    • compile a program once and inspect IR/graph + checksum,
    • run with live adapters while recording calls,
    • replay deterministically from recorded adapter traces.
  • Updated README.md and docs/DOCS_INDEX.md to expose the proof pack as the primary evidence surface for compile-once/run-many and deterministic replay.

Adapter Result Envelopes (Descriptive Contracts)

  • Extended tooling/adapter_manifest.json with result_envelope metadata for:
    • http (success envelope: ok, status_code, error, body, headers, url)
    • queue (enqueue envelope: ok, message_id, queue_name, error)
    • svc (extension/OpenClaw health envelope: ok, status, latency_ms, error)
  • Updated docs/reference/ADAPTER_REGISTRY.md to document the same envelopes and clearly mark svc as extension/OpenClaw-only (non-canonical).
  • Added tests/test_adapter_result_envelopes.py to keep the manifest and docs aligned on result-envelope field sets.

HTTP Success-Envelope Normalization

  • Normalized the runtime HTTP adapter (SimpleHttpAdapter) to return an additive success envelope on 2xx responses while preserving legacy fields:
    • keep status, body, and headers behavior unchanged,
    • add ok, status_code, error=None, and url fields on success.
  • Left non-2xx and transport failures unchanged: they still surface as AdapterError / Err and do not return a failure envelope in this pass.
  • Updated tests/test_http_adapter_contracts.py to assert the new envelope fields alongside the legacy status/body behavior.
  • Extended docs/operations/AUTONOMOUS_OPS_PLAYBOOK.md with a small, truthful snippet showing how to use the HTTP success envelope for monitoring-oriented flows.

Miscellaneous Operator-Facing Improvements

  • Updated docs/EXAMPLE_SUPPORT_MATRIX.md to tag canonical examples with ops-oriented roles (branching, resilience, webhook remediation, cron/scraper).
  • Kept all compiler and core control-flow semantics unchanged; this release is a packaging, documentation, and adapter-behavior-hardening pass for autonomous-agent use.

1.0.8-canonical-lane-classification (2026-03-09)

Canonical Lane Definition

  • Added tooling/support_matrix.json as the machine-readable support-level source for:
    • canonical vs compatible syntax intent
    • canonical vs compatible example families
    • current emitter support posture
  • Added docs/AINL_CANONICAL_CORE.md to define the recommended public AINL lane separately from the full accepted compatibility surface.
  • Added docs/EXAMPLE_SUPPORT_MATRIX.md to classify canonical and compatible repository examples for onboarding, training, and migration safety.

Documentation Wiring

  • Updated README.md to expose:
    • the canonical core doc
    • the example support matrix
    • the machine-readable support matrix
  • Updated docs/DOCS_INDEX.md with canonical-lane entry points.
  • Updated examples/README.md to reference the canonical-core and example support docs alongside the existing artifact profile contract.

Regression Safety Net

  • Added initial snapshot fixtures under tests/fixtures/snapshots/ for:
    • canonical and selected compatibility compile outputs
    • selected emitter outputs
    • selected runtime path envelopes
  • Added snapshot tests:
    • tests/test_snapshot_compile_outputs.py
    • tests/test_snapshot_emitters.py
    • tests/test_snapshot_runtime_paths.py
  • Locked graph semantic checksums for current canonical examples plus one OpenClaw compatibility example and one golden compatibility example to make future compiler extraction/refactoring parity-visible.

Warning-Only Canonical Linting

  • Added compiler-side canonical guidance warnings without changing compile success behavior.
  • Current warning-only lint coverage includes:
    • inline executable content on label declaration lines
    • split-token R request form instead of dotted adapter.VERB
    • Call without explicit ->out
    • lowercase dotted adapter verbs in non-canonical request style
    • accepted-but-noncanonical compatibility ops such as X, Loop, and cache/queue policy helpers
  • Added scripts/validate_ainl.py --lint-canonical to print warning diagnostics without failing validation.
  • Added tests/test_canonical_warning_lint.py to lock the first guidance set.
  • Added scripts/refresh_snapshot_fixtures.py to intentionally regenerate the compile/emitter/runtime snapshot baselines from current compiler/runtime behavior.

1.0.7-project-origin-attribution-hardening (2026-03-09)

Provenance and Initiator Attribution

  • Hardened public project-origin attribution across repository metadata and docs so downstream mirrors, archives, forks, and machine-indexed copies retain clear initiator evidence.
  • Recorded the human initiator explicitly as Steven Hooley with public references:
  • Updated attribution-bearing surfaces:
    • README.md
    • docs/PROJECT_ORIGIN_AND_ATTRIBUTION.md
    • docs/DOCS_INDEX.md
    • docs/PROVENANCE_AND_RELEASE_EVIDENCE.md
    • docs/GITHUB_RELEASE_CHECKLIST.md
    • CITATION.cff
    • pyproject.toml
    • NOTICE
    • tooling/project_provenance.json
  • Added emitted-artifact provenance hardening in compiler_v2.py so generated server, OpenAPI, frontend, SQL, env, deployment, and runbook outputs carry explicit project-origin references.

Public Release Positioning Polish

  • Refined the top-level README to present AINL more clearly as a graph-first AI-native language/toolchain and intermediate programming system.
  • Added a more public-facing breakdown of:
    • what AINL is,
    • what is best supported today,
    • where users should still exercise caution.
  • Updated docs/AUDIENCE_GUIDE.md with a path for first-time public evaluators.
  • Added docs/NO_BREAK_MIGRATION_PLAN.md as the operator-grade milestone tracker for compatibility-preserving canonical convergence.
  • Linked the no-break migration plan from:
    • README.md
    • docs/DOCS_INDEX.md
    • docs/POST_RELEASE_ROADMAP.md

1.0.6-runtime-compiler-contract-canonicalization (2026-03-03)

Runtime Canonical Ownership

  • Finalized canonical runtime execution ownership in runtime/engine.py (RuntimeEngine).
  • Replaced independent runtime logic in runtime.py with compatibility re-exports only.
  • Added compatibility shim runtime/compat.py:
    • preserves historical ExecutionEngine API
    • bridges legacy adapters into canonical runtime adapter interfaces.
  • Updated runtime/__init__.py exports to make canonical + compatibility surfaces explicit.

Compiler-Owned Runtime Contract Helpers

  • Added compiler-owned helpers in compiler_v2.py and switched runtime to consume them:
    • runtime_normalize_label_id()
    • runtime_normalize_node_id()
    • runtime_canonicalize_r_step()
  • Reduced runtime duplication for label/node normalization and R-step shape interpretation.

Runtime/Compiler Semantic Alignment

  • Hardened step/graph parity around:
    • Call out-binding and _call_result compatibility behavior
    • Err/Retry explicit @nX targeting
    • Loop/While deterministic behavior and limits
    • capability ops (CacheGet, CacheSet, QueuePut, Tx, Enf)
  • Broadened retry applicability and aligned strict graph-port validation so retry/error ports are accepted on executable source ops (not only R), matching runtime behavior.

Strict Dataflow Model Hardening (Compiler-Owned)

  • Fixed static-analysis gap for Call ->out ... Retry @nX ... J out strict paths.
  • Improved compiler read/write analysis (_analyze_step_rw) to reflect runtime semantics:
    • explicit Call writes
    • literal-aware read filtering in key ops.
  • Added strict quoted-literal policy enforcement via compile/dataflow behavior:
    • in strict mode, bare identifier-like tokens in read positions are treated as variable references
    • string literals must be quoted.
  • Added explicit strict error guidance to quote literals when undefined-var failures indicate likely literal intent.

Test Coverage Expansion

  • Added/expanded runtime/compiler contract tests:
    • tests/test_runtime_compiler_conformance.py
    • updates in tests/test_runtime_basic.py
    • updates in tests/test_runtime_graph_only.py
    • updates in tests/test_runtime_parity.py
    • fixture additions in tests/conformance_runtime/fixtures/retry_at_node.json
  • Added strict matrix coverage for quoted-vs-bare behavior in:
    • Set.ref
    • Filt.value
    • CacheGet.key
    • CacheGet.fallback
    • CacheSet.value
    • QueuePut.value.

Documentation and Cross-Linking

  • Added runtime/compiler contract document: docs/RUNTIME_COMPILER_CONTRACT.md.
  • Updated cross-links and status docs to reflect canonical runtime ownership and strict literal policy:
    • README.md
    • docs/DOCS_INDEX.md
    • docs/architecture/ARCHITECTURE_OVERVIEW.md
    • docs/CONFORMANCE.md
    • docs/language/grammar.md
    • docs/RELEASE_READINESS.md
    • docs/AI_AGENT_CONTINUITY.md
    • docs/CONTRIBUTING_AI_AGENTS.md
  • Added long-term docs governance contract: docs/DOCS_MAINTENANCE.md.
  • Expanded schema/profile cross-linking for small and large agents:
    • docs/reference/IR_SCHEMA.md
    • docs/reference/GRAPH_SCHEMA.md
    • docs/reference/AINL_V0_9_PROFILE.md
    • docs/PATTERNS.md
    • docs/reference/TOOL_API.md
    • docs/INSTALL.md
    • docs/language/AINL_CORE_AND_MODULES.md
    • docs/language/AINL_EXTENSIONS.md
  • Added docs automation and governance guardrails:
    • scripts/check_docs_contracts.py (stale-phrase, link-style, required-link, semantics-doc-coupling checks)
    • ainl-check-docs CLI entrypoint
    • .venv/bin/python scripts/run_test_profiles.py --profile docs
    • .pre-commit-config.yaml local hook (ainl-docs-contract) for pre-push/pre-commit parity with CI
    • CI job .github/workflows/ci.yml -> docs-contract
    • PR checklist updates in .github/pull_request_template.md

1.0.5-grammar-runtime-contract-hardening (2026-03-03)

Compiler-Owned Prefix/Grammar Contract

  • Moved remaining decoding transition ownership to compiler helpers in compiler_v2.py:
    • grammar_scan_lexical_prefix_state
    • grammar_next_slot_classes
    • grammar_prefix_line_ok
    • grammar_apply_candidate_to_prefix
    • grammar_active_label_scope
    • grammar_prefix_completable
  • Kept compiler_grammar.py as formal-only orchestration (state + admissibility).
  • Isolated non-authoritative sampling into grammar_priors.py.
  • Kept compatibility composition in grammar_constraint.py.

Formal/Prior Layering Tightening

  • Removed formal matcher fallback to prior samples; formal class matching is compiler-owned only.
  • Reworked priors to consume formal state/classes from callers (no reach-back imports into formal core).
  • Added typed protocol contracts for prior-state input to reduce silent drift risk.
  • Added compiler-owned constants for decoder control classes (NEWLINE, QUOTE_CLOSE, etc.).

Conformance Test Expansion

  • Added corpus-driven transition tests for prefix + candidate -> next prefix invariants.
  • Added corpus-wide class/sample/mask round-trip checks ensuring surviving candidates keep prefixes completable.
  • Added structural-vs-strict boundary tests to preserve distinction between prefix plausibility and strict compile validity.
  • Added runtime/compiler conformance tests to validate execution against compiler-emitted step schema for:
    • Call out-binding
    • If, Err, Retry, CacheGet, CacheSet, QueuePut, Tx, Enf

Documentation and Cross-Linking

  • Added/updated ownership contract references across:
    • README.md
    • docs/DOCS_INDEX.md
    • docs/architecture/ARCHITECTURE_OVERVIEW.md
    • docs/RUNTIME_COMPILER_CONTRACT.md
    • docs/RELEASE_READINESS.md
    • docs/CONTRIBUTING_AI_AGENTS.md
    • docs/AI_AGENT_CONTINUITY.md

1.0.4-documentation-timeline-correction (2026-03-03)

Historical Provenance Correction

  • Corrected prior timeline wording that implied AINL-origin work began in early 2025.
  • Updated project chronology to reflect:
    • 2024 foundational AI research and cross-platform experimentation by the human founder.
    • Partial loss of early artifacts, followed by explicit rebuild/retest/revalidation.
    • 2025-2026 formalization, implementation expansion, and release hardening of AINL.
  • Standardized this corrected timeline anchor across project-facing docs:
    • README.md
    • CONTRIBUTING.md
    • docs/AUDIENCE_GUIDE.md
    • docs/architecture/ARCHITECTURE_OVERVIEW.md
    • docs/RELEASE_READINESS.md
    • docs/CONFORMANCE.md
    • docs/AINL_SPEC.md
    • docs/PROJECT_ORIGIN_AND_ATTRIBUTION.md
    • CITATION.cff

1.0.3-documentation-timeline-clarity (2026-03-03)

Note: historical start-date wording in this entry is superseded and corrected by 1.0.4-documentation-timeline-correction.

Project Timeline Clarification

  • Documented an explicit historical timeline confirming that AINL research and project initiation began in early 2025.
  • Added phase framing to reduce ambiguity about when work occurred:
    • Early 2025: concept definition, AI-native language research, and naming/design exploration.
    • Mid 2025: grammar/spec experimentation, compiler direction setting, and first IR-shape validation loops.
    • Late 2025: implementation expansion across compiler/runtime paths, emitter surface growth, and conformance-first test structuring.
    • Early 2026: runtime/platform hardening, adapter contract coverage, alignment/eval gate maturity, and publication/operations documentation.
  • Added explicit note that phases overlap by design and were developed iteratively in parallel tracks (research, development, and implementation were not strictly linear).
  • Updated cross-reference documentation (README.md, docs/PROJECT_ORIGIN_AND_ATTRIBUTION.md, docs/DOCS_INDEX.md) so timeline context is discoverable from both onboarding and release-history entry points.

1.0.2-alignment-docs-and-gates (2026-03-03)

Evaluation and Alignment Pipeline

  • Added hard trend quality gates with threshold/regression checks and non-zero failure mode in scripts/analyze_eval_trends.py.
  • Added machine-readable run health output corpus/curated/alignment_run_health.json in scripts/run_alignment_cycle.sh.
  • Added prompt-length bucketing support across eval/sweep/cycle for shape-stability and better diagnostics.
  • Added optional quantized eval/infer lane (--quantization-mode none|dynamic-int8) with safe fallback behavior.
  • Added bounded host-side canonicalization controls:
    • --canonicalize-chunk-lines
    • --canonicalize-max-lines

Diagnostics

  • Added per-length-bucket diagnostics in model eval output:
    • rates, timing totals, failure families, and per-bucket constraint health.
  • Extended trend output with bucket-level summary pointers:
    • worst strict bucket
    • slowest bucket
    • quantization metadata

Documentation

  • Added docs/DOCS_INDEX.md as a top-level orientation map.
  • Added docs/AI_AGENT_CONTINUITY.md for handoff and persistence protocol.
  • Added docs/TRAINING_ALIGNMENT_RUNBOOK.md for full train/sweep/gate operations.
  • Added publication-layer docs:
    • docs/architecture/ARCHITECTURE_OVERVIEW.md
    • docs/CONTRIBUTING_AI_AGENTS.md
    • docs/reference/GLOSSARY.md
    • docs/PROJECT_ORIGIN_AND_ATTRIBUTION.md
  • Added OpenClaw integration guide: docs/adapters/OPENCLAW_ADAPTERS.md
  • Added consultant report template and index:
    • AI_CONSULTANT_REPORT_TEMPLATE.md
    • CONSULTANT_REPORTS.md
    • AI_CONSULTANT_REPORT_APOLLO.md (Apollo's OpenClaw integration analysis)
  • Added OpenClaw agent quickstart: AI_AGENT_QUICKSTART_OPENCLAW.md
  • Updated docs/FINETUNE_GUIDE.md to link and use the new runbook.
  • Added publication/trust/community artifacts:
    • docs/AUDIENCE_GUIDE.md
    • docs/GITHUB_RELEASE_CHECKLIST.md
    • CODE_OF_CONDUCT.md
    • SECURITY.md
    • CITATION.cff
    • .github/ISSUE_TEMPLATE/bug_report.yml
    • .github/ISSUE_TEMPLATE/feature_request.yml
    • .github/pull_request_template.md
  • Added Option C legal/package baseline files:
    • LICENSE (Apache-2.0)
    • LICENSE.docs
    • COMMERCIAL.md
    • TRADEMARKS.md
    • MODEL_LICENSE.md
    • NOTICE
  • Updated README.md, CONTRIBUTING.md, docs/DOCS_INDEX.md, and CITATION.cff to reflect open-core licensing structure and DCO expectations.

1.0.1-runtime-platform (2026-02-22)

Runtime

  • Added execution mode policy controls: graph-preferred, steps-only, and graph-only.
  • Added unknown-op policy controls: skip and error, with explicit behavior in both step and graph execution.
  • Added IR/version compatibility guard: runtime validates ir_version major compatibility.
  • Added runtime metadata and ergonomics:
    • RuntimeEngine.run(...) convenience wrapper
    • runtime_version in run payloads
    • trace_sink callback for streaming trace events
    • lineno included in trace events
  • Added production guardrails:
    • max_steps
    • max_depth
    • max_adapter_calls
    • max_time_ms
    • max_frame_bytes
    • max_loop_iters (applies across loop paths)
  • Hardened frame semantics and variable validation:
    • explicit failures for missing destination variables in Set, Filt, Sort, and X.
  • Refactored runtime internals to reduce semantic drift:
    • shared step execution helpers (_exec_step and common op helpers)
    • shared graph error-routing helper for err edge + active handler fallback + recursion guard.
  • Improved graph traversal performance with indexed edge lookups for common (from, port, to_kind) access.

CLI

  • Extended ainl run with:
    • execution mode and unknown-op policy flags
    • guardrail limit flags
    • --trace-out to write trace JSON
    • --record-adapters and --replay-adapters for deterministic adapter replay workflows.
  • Added pretty runtime error formatting with source snippet + caret output in non-JSON mode.
  • Added ainl golden command for fixture-driven verification using examples/*.ainl + *.expected.json.
  • Enhanced ainl check output with ir_version and compiler warnings.
  • Added direct adapter bootstrapping flags to ainl run for http, sqlite, fs, tools, and ext.

Adapters

  • Added SimpleHttpAdapter (runtime/adapters/http.py) with:
    • method allowlist
    • timeout handling
    • host allowlist
    • JSON/text request-response handling
    • response-size guard
    • consistent AdapterError mapping for transport/status/validation failures.
  • Added replay/recording adapter registries in runtime package:
    • RecordingAdapterRegistry
    • ReplayAdapterRegistry
  • Added SimpleSqliteAdapter (runtime/adapters/sqlite.py) with query/execute contract, allow-write policy, and table allowlist support.
  • Added SandboxedFileSystemAdapter (runtime/adapters/fs.py) with sandbox-root confinement, extension policy, and size caps.
  • Added ToolBridgeAdapter (runtime/adapters/tools.py) for tool-call bridging with tool allowlist and error mapping.
  • Added HttpAdapter support in adapter base and registry accessor (get_http()).
  • Added registry accessors for sqlite, fs, and tools.

Tests

  • Added property/fuzz suites (Hypothesis):
    • step-vs-graph equivalence
    • randomized IR safety checks.
  • Added runtime guardrail tests (max_* limits).
  • Added HTTP adapter contract tests (validation, timeout, error mapping, request/response shape).
  • Added deterministic replay test (live run vs replay output/log parity).
  • Added runtime API/compat tests:
    • ir_version/runtime_policy presence
    • unsupported IR rejection
    • unknown-op policy enforcement
    • trace sink and wrapper behavior
    • CLI golden pass.
  • Added runtime conformance fixture harness with trace-signature expectations.
  • Updated conformance assertions for endpoint layout compatibility.

Documentation

  • Added/updated:
    • SEMANTICS.md (frozen runtime semantics)
    • docs/RELEASE_READINESS.md (capability-to-test/file handoff map)
    • docs/INSTALL.md (runtime adapter CLI examples, record/replay usage)
    • README runtime/replay/golden usage notes.

Runner Service

  • Added deployable runtime runner service:
    • scripts/runtime_runner_service.py
    • endpoints: /run, /enqueue, /result/{id}, /health, /ready, /metrics
    • compile cache + async job worker + structured logs + trace IDs
  • Added deployment artifacts:
    • services/runtime_runner/Dockerfile
    • services/runtime_runner/docker-compose.yml
  • Added service test coverage:
    • tests/test_runner_service.py

Verification

  • Consolidated matrix result after this release set:
    • 68 passed / 0 failed
    • Property + runtime + adapter contract + replay + compat + conformance suites all green.