← ClaudeAtlas

record-project-evidencelisted

Append project claims with reproducible local evidence, artifact references, validation results, and limitations. Use when asked to capture proof, preserve an audit trail, or verify a lot during or after work without conflating plans, commands, and observed outcomes.
kotnisofiane-bit/dubsar-agent-skills · ★ 0 · AI & Automation · score 68
Install: claude install-skill kotnisofiane-bit/dubsar-agent-skills
# Record Project Evidence ## Objective Maintain an append-only evidence index that distinguishes supported facts from reported or unverified claims. ## Inputs - the shared `mission_id` and an existing `lot_id`; - one bounded claim; - inspected artifact paths, validation method and result, relevant exit status, limitations, and failed checks. ## Evidence classes - `observed`: directly inspected output or artifact; - `reported`: a statement supplied by a person or another tool; - `derived`: a deterministic calculation from cited inputs; - `unverified`: a claim that still needs proof. An executed command is not proof of success unless its result and relevant exit status were observed. A unit test does not prove an end-user workflow unless it actually exercises that workflow. ## Workflow 1. Record the claim in plain language. 2. Assign the correct evidence class. 3. Cite relative artifact paths or a concise observation reference. 4. Record the validation method and result separately. 5. Add limitations and failed checks. 6. Append a new entry; do not silently replace an earlier one. ## Output Append to `evidence.json`: ```json { "format": "dubsar.project-evidence/1", "mission_id": "same-as-mission", "entries": [ { "evidence_id": "local-stable-id", "lot_id": "related-lot", "claim": "one bounded claim", "class": "observed", "artifact_refs": ["relative/path/to/artifact"], "validation": ["method and observed result"],