Refactor Sage
Safe refactors with behavior locks
Plans refactors as behavior-preserving steps, each verifiable by existing tests, and refuses steps it cannot prove safe.
Works with: OpenClaw · Claude Code · Cursor · Any MCP agent · v1.2.0 · Apache-2.0 · by skillforge-curated
SKILL.md preview
# Refactor Sage ## Triggers User wants cleaner code without breaking behavior. ## Workflow 1. List smells with file:line evidence. 2. Propose the smallest step sequence that removes them. 3. State how each step is verified (tests/typecheck). 4. Refuse steps with no verification path. ## Guardrails - No behavior change without explicit approval. - One concern per step.
Install in one command
1. Create the skill folder: mkdir -p skills/refactor-sage
2. Install: curl -fsSL "https://skillforge-aniruddha-adaks-projects.vercel.app/api/skills/refactor-sage/skill.md" -o skills/refactor-sage/SKILL.md
3. Mention it in SOUL.md or BOOTSTRAP.md so OpenClaw loads it on boot.
curl -fsSL "https://skillforge-aniruddha-adaks-projects.vercel.app/api/skills/refactor-sage/skill.md" -o skills/refactor-sage/SKILL.md
Why this score
3/3 required SKILL.md sections, 374 chars
3 declared capabilities
Works with openclaw, claude-code, cursor, generic-mcp
Apache-2.0 — reuse freely
9400 registry installs
No risky capabilities declared
Audit seal
SHA-384(prev ‖ canonicalJson)
b365047c542b5c997c9a185ee27d7a611b57f389004a7319013eac4174be37404466aa740b602878dc7c2a84e200a0c5
chain head e800631e36483c866a9a2ba7ae87b272…
Replay: POST {ids} to /api/seals/verify