v3.5.0-beta

Highlights

This release introduces the chunk-input command for deterministic input packaging in phased execution plans, a data-loss fix for backup handling, and project-level extensibility via manifest v2.

New Features

cpt chunk-input command (#137)

New CLI command that splits oversized workflow input into line-bounded Markdown chunks for phased execution plans.

• Deterministic chunking with configurable line limits
• Dry-run mode for input signature computation without writing files
• Atomic package staging with stdin integration
• Output-dir non-directory guard in both write and dry-run modes

Input signature reuse (#137)

Plan metadata now tracks an input_signature field. When an existing input package matches the current signature, it is reused automatically — avoiding redundant re-packaging.

• input_signature and input_manifest fields added to plan.toml frontmatter schema
• Plan decomposition checks signature against existing manifest before redirecting to direct workflow
• Generate workflow performs non-mutating signature check and reuses existing packages

Project-level extensibility (#108)

Manifest v2 with multi-layer pipeline enables project-level customization of agents, skills, and workflows through layered configuration.

• Multi-layer discovery pipeline for agent and skill resolution
• Extended manifest schema supporting project-level overrides
• Layer discovery utilities for hierarchical configuration merging

Stats

• PRs merged: 2
• Files changed: 61
• Lines added: ~14,250
• Lines removed: ~205

Full Changelog: v3.4.0-beta...v3.5.0-beta

v3.4.0-beta

Cypilot v3.4.0-beta — RalphEx Delegation, Native Phase Agents, Plan Brief-First Flow

New Commands

cpt delegate <plan_dir>

Compiles a Cypilot plan into ralphex-compatible Markdown and delegates execution to RalphEx.

• Three modes: --mode execute (full autonomous run), --mode tasks-only (assemble without invoking), --mode review (read-only analysis)
• Auto-discovers ralphex on PATH or via persisted core.toml config
• Bootstrap gate — missing .ralphex/config blocks with setup guidance
• --dry-run assembles the command without invoking ralphex
• --worktree requests worktree isolation for execute/tasks-only modes

cpt doctor

Environment health check. Currently validates ralphex availability with PASS/WARN/FAIL results. Supports --json output.

New Subagents

• cypilot-ralphex — manages discovery → export → delegation → handoff when ralphex is available
• cypilot-phase-compiler — compiles exactly one plan phase from its brief in an isolated agent context
• cypilot-phase-runner — executes the next phase from a generated plan inside a dedicated agent context

All three registered in agents.toml and generated for Claude Code, Cursor, Copilot, and Codex via cpt generate-agents.

Plan Workflow: Brief-First Flow

/cypilot-plan now pauses after writing the manifest and briefs, presenting a 4-way choice:

1. Compile phases inline in the current chat
2. Emit self-contained per-brief prompts for downstream workers
3. Run cypilot-phase-compiler subagents
4. Stop here — keep manifest + briefs without compiling

This replaces the previous auto-compile flow and gives full control over phase production strategy.

Bug Fixes

• Codex agent generation (#125) — cpt generate-agents for OpenAI Codex generated only 1 skill file instead of 5, and the combined .toml format caused Codex CLI warnings ("must define developer_instructions"). Now generates one .toml per agent with proper top-level fields.
• Workspace git timeouts — cpt workspace-sync timed out on slow networks (hardcoded 120s). Default increased to 300s; override via GIT_TIMEOUT env var.
• Zip-slip in kit download — cpt kit install from GitHub didn't validate tarfile member paths, allowing theoretical path traversal. Now validates before extraction.
• Silent error swallowing — ~15 places in kit config readers, agent discovery, and migration helpers used broad except Exception, hiding unexpected errors. Replaced with specific exception types (OSError, ValueError, tomllib.TOMLDecodeError) so real bugs surface.

Documentation

• New guides/CONFIGURATION.md — comprehensive guide covering core.toml, artifacts.toml, rules, WHEN-rules, @cpt-* traceability, and copy-pastable prompts for every configuration use case
• README — added Configuration section with file reference table and link to the guide

Breaking Changes

None. All new features are additive.

Upgrade

cpt update

Or from source:

cpt update --source . --force

---

Merged Pull Requests

• Enable Pylint CI by @ainetx in #127
• fix(workspace): increase default git timeouts, support env var for ov… by @nonameffh in #129
• docs(architecture): add ADR-0018 for RalphX delegation skill by @ainetx in #128
• ci(sonarqube): add SonarQube integration with coverage reporting by @ainetx in #130
• chore: remediate high-risk pylint warnings across cypilot by @ainetx in #131
• fix(agents): codex gets full skill set and valid per-agent TOML by @bsacrobatix in #135
• feat: add ralphex delegation flow and native phase agents by @ainetx in #134

Full Changelog: v3.3.0-beta...v3.4.0-beta

v3.3.0-beta

What's New in v3.3.0-beta

Summary

Authoritative kit paths, bug-finding methodologies, and workflow hardening.

Highlights

• Authoritative installed kit roots

  • Kit install, update, migrate, and validate flows now consistently treat config/core.toml kit paths as the source of truth.
  • Custom kit roots are preserved.
  • Absolute-path handling is safer, including inaccessible paths and cross-platform absolute path semantics.

• Manifest and migration error hardening

  • Binding-resolution failures now surface explicitly.
  • Rollback and cleanup paths are tighter.
  • Kit and manifest validation report invalid path scenarios more deterministically.

• New bug-finding methodologies

  • Added compact, high-recall requirements/bug-finding.md for source-code review.
  • Added requirements/prompt-bug-finding.md for prompt and instruction review.
  • Both are integrated into workflows for bounded, evidence-driven analysis.

• Workflow and prompt-quality tightening

  • SKILL.md, workflows/analyze.md, workflows/generate.md, and workflows/plan.md now better enforce:
    • task-matched dependency loading
    • compact prompt context
    • terminal response-completion gates
  • This reduces premature completion and improves execution discipline.

Full Details

This release improves correctness and resilience in kit path resolution, adds dedicated methodologies for bug-finding in both code and prompts, and hardens the main workflow/prompt stack to better control context loading and completion behavior.

Full Changelog: v3.2.2-beta...v3.3.0-beta

v3.2.2-beta

What's Changed

• refactor(update): extract whatsnew handling from kit update flow by @ainetx in #121

Full Changelog: v3.2.1-beta...v3.2.2-beta

v3.2.1-beta

What's Changed

• refactor(init): unify root AGENTS.md and CLAUDE.md managed blocks by @ainetx in #120

Full Changelog: v3.2.0-beta...v3.2.1-beta

v3.2.0-beta

What's Changed

• feat: ship prompt compactification across workflows, requirements, and SKILL by @ainetx in #119

Full Changelog: v3.1.0-beta...v3.2.0-beta

v3.1.0-beta

What's Changed

• feat(agents): add subagent registration for multi-tool IDE integration by @bsacrobatix in #105
• feat(requirements): add plan validation checklist and compilation brief template by @ainetx in #117

Full Changelog: v3.0.15-beta...v3.1.0-beta

v3.0.15-beta

What's Changed

• fix(utils): sanitize comment to avoid false positive from secret scan… by @enjiruuuu in #114
• fix(kit): register new manifest resources in core.toml during kit update by @ainetx in #115

New Contributors

• @enjiruuuu made their first contribution in #114

Full Changelog: v3.0.14-beta...v3.0.15-beta

v3.0.14-beta

What's Changed

• feat(skill): prefer cpt CLI over direct script invocation for AI agents by @bsacrobatix in #107
• fix(manifest): avoid high-entropy string literal in _ID_CHARS by @cashmisa in #109
• feat: add multi-repo workspace federation support by @nonameffh in #81
• refactor(architecture): reorganize .core structure, move ADRs and spe… by @ainetx in #112

New Contributors

• @bsacrobatix made their first contribution in #107
• @cashmisa made their first contribution in #109
• @nonameffh made their first contribution in #81

Full Changelog: v3.0.13-beta...v3.0.14-beta

v3.0.13-beta

What's Changed

• feat(plans): execution plans workflow, plan escalation gate, routing fix by @ainetx in #106

Full Changelog: v3.0.12-beta...v3.0.13-beta