← ClaudeAtlas

loomlisted

Extract requirements from conversations, link to code, detect drift, decompose specs into atomic tasks, and run those tasks on a local small model. Use when making decisions, before modifying code, when a spec is ready for implementation, or to check staleness.
jsuppe/loom · ★ 3 · AI & Automation · score 71
Install: claude install-skill jsuppe/loom
# Loom 🧵 — Requirements Traceability & Small-Model Task Execution Skill **Weaving requirements through code — and driving small-model code execution with them.** Loom captures decisions as versioned requirements, expands them into specifications, links code back to them with content hashes for drift detection, and (as of the `Task` + `loom decompose` + `loom_exec` layer) decomposes specs into atomic, executor-ready tasks that a local small model can complete against a full context bundle. ## When to use This skill is always active via AGENTS.md integration. Invoke at these moments: | Moment | Action | Command | |--------------------------------------|-----------------------------------------|----------------------------------| | Onboarding a new target repo | Write `.loom-config.json` + health-check| `loom init` | | Decision made in chat | Extract requirement (with rationale) | `loom extract --rationale ...` | | Before modifying code (automatic) | Pre-edit briefing | `hooks/loom_pretool.py` (hook) | | Manual drift check | Inspect a file | `loom check <file>` | | After implementing | Link to reqs or specs | `loom link <file> --req/--spec` | | Spec ready for implementation | Decompose into atomic tasks