How ericrisco/rsc-harness works
rsc-harness is a CLI product for selecting, installing, and maintaining project-specific AI-agent harnesses. Its principal workflow starts with user goals and project evidence, produces a reviewable plan, then materializes the accepted skills, agents, rules, and target-specific files and verifies the result. The README also documents lane-based work guidance and self-recommending growth. The graph preserves those capabilities; several implementation details are represented at their source-directory level because their source was not sampled.
Main entry points: install-apply.js Harness installer · onboarding-apply.js Plan and consent · rsc.js rsc CLI · manifest-file.js Project declaration
Components 22
Planning and catalog
- rsc CLI
scripts/rsc.js - Project scan
scripts/lib/onboarding.js - Skill catalog
scripts/lib/manifest.js - Catalog search
scripts/consult.js - Skill recommendations
scripts/lib/recommend.js - Plan and consent
scripts/lib/onboarding-apply.js
Installation and verification
- Harness installer
scripts/install-apply.js - Target adapters
targets/index.js - Agent definitions
targets/agents.js - Project declaration
scripts/lib/manifest-file.js - Local install state
scripts/lib/state.js - Install diagnostics
scripts/doctor.js
Harness capabilities
- Lane workflows
scripts/spec-gate.js - Agent commands
targets/commands.js
Memory and maintenance
- Project memory
targets/memory.js - Session memory
targets/session-memory-core.mjs - Worktree cleanup
targets/worktree-reaper.mjs - Turn lane gate
targets/sello.mjs
Skill tools
- Prompt extraction
skills/roast-me/tools/extract_prompts.py - Runtime adapters
skills/roast-me/tools/adapters/base.py
Other components
- Project owner
- Equipped project
Connections 23
- Project owner → rsc CLI: requests setup
- rsc CLI → Plan and consent: starts onboarding
scripts/rsc.js - Plan and consent → Project scan: reads project
scripts/lib/onboarding.js - Plan and consent → Skill catalog: loads skills
- Plan and consent → Harness installer: applies accepted plan
scripts/lib/onboarding-apply.js - Project owner → Plan and consent: accepts plan
- rsc CLI → Catalog search: searches catalog
scripts/rsc.js - rsc CLI → Skill recommendations: suggests skills
scripts/rsc.js - Skill recommendations → Skill catalog: uses skill data
- Harness installer → Target adapters: materializes targets
scripts/install-apply.js - Harness installer → Agent definitions: reconciles agents
scripts/install-apply.js - Harness installer → Project declaration: reads declaration
scripts/install-apply.js - Harness installer → Local install state: reads install state
scripts/install-apply.js - Target adapters → Project declaration: reads declaration
targets/index.js - Install diagnostics → Local install state: checks state
scripts/doctor.js - Install diagnostics → Project declaration: checks declaration
scripts/doctor.js - Equipped project → Lane workflows: uses installed lanes
- Lane workflows → Agent commands: offers repeatable workflows
- Equipped project → Project memory: uses project memory
- Equipped project → Session memory: captures sessions
- Equipped project → Turn lane gate: classifies each turn
- Equipped project → Worktree cleanup: retires merged worktrees
- Prompt extraction → Runtime adapters: parses runtimes
skills/roast-me/tools/extract_prompts.py