Synapse Engine v2
- Production-hardened
synapse_engine_v2.pywith async trigger firing - Formal decision types:
halt,warn,allow - Wired into pipeline execution at pre-execution phase
- Per-step synapse validation with blocking decisions
Complete Synapse Module (src/archon/synapses/)
- All 12 synapses now fully operational: anti_rationalization, code_quality, completeness, consistency, metacognition, mcp_client, pattern_recognition, security_awareness, sequential_thinking, trust_verification, code_quality_mcp, security_awareness_mcp
metacognition.pyexpanded from 6-line stub to 108-line full validator (plan, reasoning, reflection, confidence checks)trust_verification.pycontradiction detection fixed (was silent pass)synapses/__init__.pyexports all 12 (was only 3)
Testing Enforcement
- CI now runs
pytest tests/— 362 test suite is guarded on every commit - Added
[project.optional-dependencies][dev]with pytest, pytest-asyncio, pytest-cov - New test suite: test_anticrationalization, test_code_quality, test_completeness, test_consistency, test_metacognition, test_pattern_recognition, test_security_awareness, test_trust_verification, test_synapses_executable
MCP Servers
servers/file-ops/— File operations MCP server (Python)servers/forge/— Forge MCP serverfile-ops-rs/— Rust daemon with rate limiting + metrics
Quality
- All build artifacts properly gitignored
.gitignoreentries for file-ops-rs/target/ and servers/forge/target/- Comprehensive release notes and changelog
All notable changes to Archon will be documented in this file.
The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.
- Claude Code Hooks Architecture — 8 lifecycle hooks that transform Archon's 5 synapses from passive documents into active enforced behavior
session_boot.py(SessionStart): Project detection, git state, boot context injectionprompt_router.py(UserPromptSubmit): Complexity classification, skill routing, synapse activationguard_bash.py(PreToolUse/Bash): Blocks dangerous commands (rm -rf, force push, fork bombs)guard_write.py(PreToolUse/Write|Edit|NotebookEdit): Secret scanning with env-var false-positive preventionquality_write.py(PostToolUse/Write|Edit|NotebookEdit): Tracks modified files in session statequality_bash.py(PostToolUse/Bash): Tracks test/build pass/fail resultscompletion_gate.py(Stop): Blocks session completion when tests or build failedagent_context.py(SubagentStart): Role-specific context slicing for subagents
- Shared modules for hooks:
state.py(atomic JSON state),classifier.py(complexity tiers),scanner.py(secret/command detection) - archon-adversarial-review skill — PRISM-A protocol for independent framework auditing with 9 probe vectors
- Stripped multi-platform support — Archon is now Claude Code only (model floor: Opus 4.6)
- Removed platform-specific adapters (Copilot CLI, Cursor, Windsurf, Antigravity)
Archon v1.0.0 is the Virtuoso Engine for AI Agents — Claude Code only.
Core Framework
- 98 universal skills in a standardized format (SKILL.md + manifest.yaml + resources)
- 16 domain bundles (Godot, Web Dev, UX, Django, SDD, Testing, Mobile, Meta, Prompts, Security, Data Layer, DevOps, Windows, Orchestration, GitHub, Teaching)
- 17 agents with personas, skill bindings, guardrails, and handoff protocols
- 8 resumable multi-agent pipelines with failure recovery and context curation
- 5 cognitive synapses (metacognition, anti-rationalization, sequential-thinking, security-awareness, pattern-recognition)
Architecture
- 6-layer architecture: Skills -> Agents -> Synapses -> Pipelines -> Guardrails -> Runtime
- Session state machine with policy engine
- MCP agent-router for on-demand agent discovery
Platform
- Claude Code (model floor: Opus 4.6)
- Python SDK with programmatic access to all framework capabilities