← ClaudeAtlas

loom-evidencelisted

Use when observations, validation outputs, reproductions, logs, screenshots, scans, command results, or artifact pointers should remain available for review or closure claims.
z3z1ma/agent-loom · ★ 15 · Code & Development · score 77
Install: claude install-skill z3z1ma/agent-loom
# loom-evidence Evidence is Loom's observation surface. It records what was seen, how it was seen, source or record state, which stable claims the observation supports or challenges when applicable, and what the observation does not show. Evidence gives tickets, audit, research, specs, plans, and knowledge honest facts to reason from. It does not decide acceptance, intended behavior, policy, audit verdicts, or closure. ## Use This Skill When Use this skill when an observation should survive the session: tests, command results, inspections, reproductions, logs, screenshots, scans, artifact pointers, validation outputs, or any fact a review or closure claim will depend on. Small local checks can stay in a ticket journal when durable inspection is not needed. ## Dispatch If creating evidence: - read `references/creating-evidence.md` - read `references/evidence-quality.md` - use `templates/observation.md` for one observation - use `templates/dossier.md` when multiple observations compose one validation story - link related records only when useful If updating evidence: - preserve the original observation for single-observation records - add clarification, limits, freshness, supersession, or related-record notes - create a new record for a new observation unless a dossier is meant to collect that validation story If finding or summarizing evidence, inspect `.loom/evidence/` and keep observation, inference, support, challenge, and acceptance distinct. ## Finding Ev