executing-plans
executing-plans
Use when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is available
factory-collectExperimental workflow for collecting and triaging product feedback, product telemetry, runtime errors, and issue reports. Use when scanning configured signal sources and applying separate fix, reply, or close rules.
Independent GitHub snapshot · Automated classification · Original documentation governs
Read .agent-factory/config.yaml and apply the optional
skill_prompts.factory-collect entry as additional project guidance. Collect
only the sources named by workflows.collect.sources. For repeated patterns
across time or sources, use factory-lookback.
empty, unavailable, and truncated distinct. A missing connector,
partial page, or failed query is not an empty source.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 retaining each source record and reporter.
Implement only when workflows.collect.implement allows the change and the
repository is in scope. Stop for configured risk conditions, unclear product
intent, or evidence that cannot be reproduced. Verify the changed behavior
with the configured checks and a representative reproduction.
| Action | Gate |
|---|---|
| Reply | When missing information blocks triage or verification, ask one targeted question only if workflows.collect.reply allows it; follow configured tone and guidance. never means no public reply. |
| Close or mark fixed | Follow workflows.collect.close and wait for its proof point. Do not imply an unverified release. |
| Publish, approve, merge, or deploy | Requires its own workflow authorization; collection does not grant it. |
When an item needs more information, keep it unresolved and retain the source
thread link and the exact question. If replies are disabled, include that
question in the report for a person to send. On later collection runs, check
whether an answer arrived. Re-triage the original report together with the
answer, then apply the implementation and verification rules again; receiving
an answer does not itself authorize a fix or external action. Use
factory-lookback to compare these follow-ups with prior reports and fixes.
Summarize coverage by source, including unavailable or truncated reads. For each item or related group, give its links, classification, evidence, disposition, checks, any external action, and the exact human decision still needed.