Part of the experimental Factory suite of builder-skills. Reads .agent-factory/config.yaml (artifact-autonomy-policy) and its own optional skill_prompts overlay, which adds project guidance but cannot grant permission. It collects only the sources named by workflows.collect.sources.

Collect and verify coverage. Resolve each source ID to the connected read tool and exact scope. Follow pagination to the end and record filters, range, cursor/page coverage, and source counts. “Keep empty, unavailable, and truncated distinct. A missing connector, partial page, or failed query is not an empty source.” For telemetry, report the metric and aggregation window, and distinguish event counts from unique users unless the source has a reliable identity definition.

Triage and act. Classify each item as a verified defect, repeated symptom, feature request, subjective feedback, duplicate, out of scope, or needing more evidence. Group related reports while keeping each source record and reporter. Implement only when workflows.collect.implement allows it, and verify with the configured checks plus a representative reproduction. Reply, close, and everything downstream have separate gates. “Publish, approve, merge, or deploy: requires its own workflow authorization; collection does not grant it.”

The follow-up loop. When one missing detail blocks triage and the reply policy allows it, ask one targeted question. If replies are disabled, put the question in the report for a person to send. Later runs check for an answer, re-triage the original report with it, and apply the fix rules again. “Receiving an answer does not itself authorize a fix or external action.”

This is the intake flavor of stage-align, the same slot as pstack-benny-triage, mp-triage, and ce-sweep. The inputs are production feedback, telemetry, and errors rather than a developer’s prompt, and it runs as a scheduled workflow with nobody watching.

See Also