Agent Process Learnings Policy

This policy is optimized for two goals:

1. Keep the shared repo context clean and high-signal.
2. Keep temporary learning capture fast for humans and agents.

Quick Rules

1. Do not commit per-task learning logs.
2. Record temporary learnings locally in tmp/agent-learnings/.
3. Promote only durable, broadly useful rules into the relevant AGENTS.md file via normal PR review.

tmp/ is gitignored in this repository, so local notes stay local by default.

What Gets Committed

Artifact	Location	Commit?	Purpose
Temporary task learnings	tmp/agent-learnings/<YYYY-MM-DD>-<topic>.md	No	Working memory during/after a task
Durable process rules	AGENTS.md (root or component)	Yes	Stable guidance for future tasks

Fast Workflow

1. Capture guidance and mistakes in a local note while working.
2. At task end, decide if any pattern is durable.
3. If durable, propose a concise AGENTS.md update (what happened, root cause, updated rule).
4. Discard or keep local notes locally; do not commit them.

Local Note Template (Copy/Paste)

# <YYYY-MM-DD> - <topic>

- Date:
- Context (task/area):
- What went wrong or where we got stuck:
- Human guidance/correction:
- Root cause:
- Proposed durable update:
  - Target file/path (usually `AGENTS.md`):
  - Rule/procedure to add or change:
- Follow-up checks: