All notable changes to this project will be documented in this file.
The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.
- Showcase #20: Agent Stock Exchange β multi-round marketplace simulation where 3 analyst agents bid on 20 deterministic tasks across 4 categories (data/web/security/infra). Exercises AgentGraph, BroadcastManager, AgentRouter, and BasicAgent + data_slush simultaneously. Emergent specialization, reputation effects, and wealth distribution.
- 5 remaining UI-called RPC methods registered in method files for MockServer/test parity
chat.messagesβ retrieve session messages with optional limitchannels.sendβ send a message via channel registryagents.files.read,agents.files.writeβ read/write agent files via registryconfig.applyβ apply raw config with configManager or in-memory fallback
- 16 new tests across
dashboard-rpc.test.tsandgateway-rpc-methods.test.ts - Total test count: 2769 tests across 106 files
- Dashboard RPC parity: All 12 UI pages now fully functional β 19 missing RPC methods registered
chat.list,chat.deleteβ session management for chat and sessions pagescron.list,cron.add,cron.enable,cron.run,cron.removeβ full CRUD for cron pageskills.list,skills.toggleβ skill listing and enable/disable for skills pageagents.listβ agent summary listing for agents pagechannels.list,channels.connect,channels.disconnect,channels.probe,channels.configureβ channel ops for channels pageconnections.listβ device listing for devices pagestatus,healthβ system info for debug and presence pages
- 3 new method files:
channels-methods.ts,connections-methods.ts,system-methods.ts dashboard-rpc.test.tsβ 30 new handler tests for all dashboard RPC methods- Updated
gateway-rpc-methods.test.tsβ 25 β 55 tests covering 18 method groups - Total test count: 2753 tests across 106 files
- Stale version references in
CLAUDE.md(1.6.0 β 1.8.0) andskills.md(1.4.0 β 1.8.0) - Empty
__init__.pyfiles in 7 Python sub-packages now have proper exports with__all__
- Export tests for all 7 Python sub-packages (
test_module_exports.py) - CHANGELOG entries for v1.5.0βv1.8.1
- Parallel AgentGraph execution in Python (
python/openrappter/agents/graph.py) - 9 Python showcase ports: Darwin's Colosseum, Infinite Regression, Ship of Theseus, Panopticon, Lazarus Loop, Agent Factory, Swarm Vote, Time Loop, Ghost Protocol
- 11 new Python modules: channels, config, gateway, mcp, memory, security, storage sub-packages
- 81 new Python tests across showcase and parity test suites
- Version bump to 1.8.1 in
package.jsonandpyproject.toml
- Python parity:
AgentChain,AgentGraph, andAgentTracerported to Python - Chat methods for gateway WebSocket protocol
- 151 new tests across TypeScript and Python
- Swift agent fixes for actor isolation
- Phoenix Protocol: Self-healing agent orchestration (32 tests)
- 19 Showcase Prompts: Advanced agent orchestration patterns with runnable examples
- The Architect, Ouroboros Accelerator, Swarm Debugger, Mirror Test, Watchmaker's Tournament
- Living Dashboard, Infinite Regression, Code Archaeologist, Agent Compiler, Doppelganger
- The Inception Stack, Data Sloshing Deep Dive, Memory Recall, Channel Switchboard
- Config Hotswap, Persistence Vault, Healing Loop, Authorization Fortress, Stream Weaver
- Showcase dashboard UI page (
<openrappter-showcase>Lit web component) - Showcase RPC methods:
showcase.list,showcase.run,showcase.runall - 176 showcase tests (all deterministic, no LLM calls)
- AgentGraph: DAG executor with parallel execution, topological sort, cycle detection, multi-upstream
data_slushmerging - AgentTracer: Span-based observability for agent execution (start/end/duration/inputs/outputs)
- MCP Server: Expose agents as Model Context Protocol tools via JSON-RPC 2.0 over stdio
- Dashboard REST API: HTTP endpoints for web dashboard (
/api/agents,/api/traces,/api/status) - Python parity tests for broadcast, router, subagent patterns
- AgentChain: Sequential pipeline with automatic
data_slushforwarding, transforms, timeouts - LearnNewAgent TypeScript port: Runtime agent generation with hot-loading, factory pattern
- LLM-powered agent description inference for LearnNewAgent
- 10 LearnNewAgent runtime generation prompts
- 10 agent chain prompts
- Single File Agent Pattern: The defining architecture of openrappter
- One file = one agent. Metadata contract, documentation, and deterministic code all in a single
.pyor.tsfile - Native code constructors: Python dicts and TypeScript objects β no YAML, no config files, no magic parsing
slush_out()(Python) /slushOut()(TypeScript) β convenience helper for buildingdata_slushdictsSubAgentManagerauto-chainsdata_slushbetween sequential sub-agent calls viacontext.lastSlushBroadcastManagerfallback mode passesdata_slushfrom failed agents to the next in the chain
- One file = one agent. Metadata contract, documentation, and deterministic code all in a single
- Single File Agent Manifesto: RappterHub page explaining the standard
- All built-in agents use the native constructor pattern
LearnNewAgentgenerates agents with native code constructors
- Data Slush: Agent-to-agent signal pipeline
- Agents can return a
data_slushdict in their JSON output with curated signals from live results last_data_slush(Python) /lastDataSlush(TypeScript) property onBasicAgentfor accessing the most recent outputupstream_slushkwarg onexecute()β automatically merged intoself.context['upstream_slush']for downstream agents- Enables LLM-free agent chaining in sub-agent pipelines, cron jobs, and broadcast patterns
- Agents can return a
WeatherPoetAgentβ example agent demonstrating data_slush with live weather API integration and haiku generationupstream_slushfield added toAgentContexttype (TypeScript)
- Monorepo structure: Separate
python/andtypescript/directories - TypeScript agent system: Full port of Python agent pattern to TypeScript
BasicAgent.tswith data sloshingAgentRegistry.tsfor dynamic agent discoveryShellAgent.tsandMemoryAgent.tscore agents
- Unified agent contract between Python and TypeScript
pyproject.tomlfor Python packaging
- Reorganized repository structure for dual-runtime maintenance
- Python package moved to
python/openrappter/ - TypeScript source moved to
typescript/src/ - Updated all documentation for monorepo structure
- Lowered Node.js requirement to 18+ (from 22+)
- Dynamic agent discovery system (agents/ directory)
- BasicAgent base class following CommunityRAPP pattern
- Data sloshing for context enrichment
- Agent switching at runtime (
/agent <name>,/agents) --list-agentsand--agentCLI options
- Renamed RAPPagent.py to openrappter.py
- Lowercase "rapp" throughout for readability
- Restructured to agents/ directory pattern
- Initial release of openrappter
- GitHub Copilot SDK integration (no API keys needed!)
- Interactive chat mode
- Single task execution (
--task) - Persistent memory system
- Built-in skills: bash, read, write, list
- Custom skill support (YAML and Python)
- Onboarding wizard
- Python standalone version (openrappter.py)
- Full documentation and GitHub Pages site
- Node.js 18+ required
- TypeScript with strict mode
- ESM modules
- Vitest for testing