retro

What it does

retro mode reads your messages from the current session and proposes up to five rules based on what you corrected or rejected, each one backed by your own words. A script checks every proposal, and only the ones you accept are saved, in modes/_custom.md or modes/_profile.md.

When to use it

Run retro at the end of a session in which you corrected career-ops, for example "stop writing my summary in third person". A correction you make in chat lasts for that session; a rule you accept here still holds in the next one. It needs you there to approve: in an unattended run it stops and asks you to run it in an interactive session.

What a run looks like

> /career-ops retro

1. It finds the lessons. A lesson is a correction or a rejection that should still hold next session.
2. It sorts them. A fact about you, such as a role, a skill or a metric, is not a rule: retro tells you so and points you to the mode that records it. A scoring or targeting preference goes to your profile, and any other standing preference goes to your custom rules.
3. It quotes you. Each proposal carries your own words, copied exactly, as its evidence.
4. It checks them. learn.mjs validates the proposals and numbers them, and retro shows you that output unchanged. A blocked proposal stays blocked; retro does not reword it to get it through.
5. You choose by number. Name the ones to accept or reject. Any number you don't name stays undecided.
6. It reports what changed. It names each file and line. If the session has no lessons, it says so and stops.

What it does not touch

retro does not edit your CV, article-digest.md or config/profile.yml. It writes two working files under data/learning/, and only learn.mjs writes anything else: the rules you accepted, a .bak copy of each file it changes, and a record of every decision in data/learning/ledger.jsonl.