Learnings

LEARNINGS

This skill manages and uses Firstpick's durable troubleshooting LEARNINGS archive.

The archive is not chat memory. It is a local, file-based knowledge base of short solved-issue notes plus generated indexes and summaries.

When to use

Use this skill when:

• a troubleshooting issue looks similar to something solved before
• the user asks about LEARNINGS, /ret-LEARNINGS, or /sum-LEARNINGS
• a troubleshooting session ended with a real solution and should be recorded
• the LEARNINGS symlink/env/timer/scripts need inspection or repair

Do not blindly apply a past solution. Always verify against current files, commands, logs, and system state.

Path resolution

Resolve the archive root in this order:

1. Source ~/.pi/agent/learnings.env if present and use $LEARNINGS_DIR.
2. Otherwise use the stable symlink ~/.pi/agent/LEARNINGS.
3. Otherwise ask before creating a new archive.

Expected current value on Firstpick's system:

/mnt/SSD_NVME/LEARNINGS

Portable shell snippet:

source "$HOME/.pi/agent/learnings.env" 2>/dev/null || true
LEARNINGS_DIR="${LEARNINGS_DIR:-$HOME/.pi/agent/LEARNINGS}"

Retrieval workflow

1. Resolve LEARNINGS_DIR.
2. Read $LEARNINGS_DIR/LEARNINGS-SUMMARY.md.
3. Identify relevant entries.
4. Read the referenced source or archive file for each relevant entry.
5. Apply the lesson only after verifying the current system/repo state.
6. In the final answer, briefly cite which LEARNINGS files influenced the solution.

If LEARNINGS-SUMMARY.md is missing or stale, run:

~/.pi/agent/bin/learnings-summary

or, from this skill directory, using paths relative to the skill root:

../../scripts/sync-learnings.py
../../scripts/summarize-learnings.py

Adding a learning

This is an implicit completion requirement, not an optional reminder.

At the end of every troubleshooting/configuration/repair task where a real solution was found, automatically create or update one concise LEARNINGS .md note before the final answer. Do not wait for the user to ask. This applies especially when this skill was used for inspection, repair, cleanup, setup, repeated-issue retrieval, or LEARNINGS maintenance.

Skip writing a note only when one of these is true:

• no solution was found
• the task was pure explanation/research with no fix or reusable operational lesson
• the user explicitly says not to write a learning
• writing would duplicate an existing note; in that case update the existing note only if the new solution materially changes it

Use this format:

# concise-title-date

- Issue: what happened.
- Tried: key diagnostics/fixes attempted.
- Solution: what worked at the end.
- Verification: command/output or observed behavior proving it worked.

If no solution was found, skip creating a note unless the user explicitly asks for a failed-attempt record.

After adding or updating a note, always regenerate summary/index:

~/.pi/agent/bin/learnings-summary

Before the final answer, verify the note exists and mention the path briefly.

Maintenance workflow

Health check:

source "$HOME/.pi/agent/learnings.env" 2>/dev/null || true
LEARNINGS_DIR="${LEARNINGS_DIR:-$HOME/.pi/agent/LEARNINGS}"
ls -la "$HOME/.pi/agent/LEARNINGS" "$HOME/.pi/agent/learnings.env" "$LEARNINGS_DIR"
python -m json.tool "$LEARNINGS_DIR/manifest.json" >/dev/null
systemctl --user list-timers sync-learnings.timer --all --no-pager

Expected core files:

README.md
INDEX.md
manifest.json
LEARNINGS-SUMMARY.md
archive/
logs/
sync-learnings.py
summarize-learnings.py
run-sync-with-notification.sh

Package scripts

This package includes maintenance scripts in scripts/:

• sync-learnings.py — archives root learning notes and writes manifest.json + INDEX.md.
• summarize-learnings.py — generates LEARNINGS-SUMMARY.md from manifest.json.
• run-sync-with-notification.sh — runs both and writes logs/latest.log.

When referenced from this skill, package-level helper scripts use explicit paths relative to this skill root, e.g. ../../scripts/sync-learnings.py.