loopany-capture

# loopany-capture — event-triggered artifact capture After substantive work concludes, check whether to record it as a loopany artifact. "Substantive" = produced evidence or a decision a future reader would need to find. ## Event → Kind routing | Event | Target kind | Read | |-------|-------------|------| | PR shipped / merged | `task` | `../loopany-core/kinds/task.md` | | Incident resolved | `task` | `../loopany-core/kinds/task.md` | | Problem noticed, not acting now | `signal` | `../loopany-core/kinds/signal.md` | | Existing task outcome observed | append `## Outcome` | `../loopany-core/kinds/task.md § Playbook` | | Decision made with rationale | `note` or `task --status done` | see flowchart below | ### Decision flowchart - Resolved a pending signal? → `task --status done` + `addressed` edge - Produced something shippable? → `task --status done` with `## Outcome` - Reasoned preference / principle? → `note` titled "decided: X over Y because Z" - Belief from ≥ 2 data points? → `learning` via `loopany-reflect` Default `note` when unsure. ## Quality gate Skip capture when: - **Not observable** — feeling, not fact. - **Already captured** — `loopany artifact list --contains "<phrase>"`. Match? Append. - **Too small** — typo fix, whitespace, one-liner. - **Speculation** — at most a `signal`, only if concrete. - **Internal agent work** — grepping, reading, answering. Not a capture event. Quick test: if you can't write a one-sentence `## Outcome`, skip the task. ## Subagent dispatch Don't capture inline — it breaks focus and produces weaker artifacts: 1. **Filter** — main session runs the quality gate. 2. **Compose context** (3-5 sentences): what happened, why it matters, any numbers or artifact IDs in scope. 3. **Dispatch**: "Record this via loopany. Read `loopany-core` to pick the right kind. Return the artifact ID." 4. Subagent writes the artifact, returns ID. 5. Main session continues — no context switch. If the host has no subagent primitive, capture inline but keep it tight. ## Anti-patterns - ❌ Trigger on every message exchange → only at completed-work boundaries. - ❌ Capture during work, not after → no Outcome yet. - ❌ Skip the quality gate → noise artifacts pollute reflect. - ❌ Inline capture in long sessions → use subagent pattern.