| name | review-learnings |
| description | Review accumulated retrospective findings in the learnings inbox and decide which to apply as guidelines, skills, architecture tests, or hooks. Use when the user wants to triage the inbox. |
| scope | argos |
Learnings Review
1. Read the inbox
cat .ai/learnings/inbox.md
If empty or missing, report: "No pending learnings to review."
The inbox is written in German — read and discuss accordingly. The user may
converse in German or English at their preference.
2. Process each finding
For each finding, present it with three options:
- Apply → integrate at the proposed location (guideline / skill /
architecture test / hook)
- Defer → keep in inbox, decide later
- Discard → noise, not worth acting on
For "Apply":
- Draft the actual change (new lines in guideline, new skill file,
new arch test, new hook entry) — in English, regardless of inbox
language. The artifact going into the codebase is always English.
- Show the diff.
- Get explicit confirmation.
- Write the change.
- Mark the finding as processed (move to a "processed" section, do not
delete yet — keep traceability until archive).
NEVER apply changes without per-finding confirmation. Do not batch-apply.
3. Archive processed findings
When all findings in the current inbox are processed, move the file to:
.ai/learnings/archive/YYYY-MM-DD-HHmm.md
(Use the date of the review, not the original finding dates.)
Creating a fresh empty inbox is not required — the next /retro will
create one if missing.
4. Summary
Report:
- N findings processed
- M applied (with one-line each), K deferred, L discarded
- Inbox archived to
<path>
