Naval Save Learning

Save only the reusable insight, not a transcript or generic activity log.

Read First

• ../../references/memory/README.md
• ../../references/memory/templates/learning.md
• ../../references/memory/schemas/learning.yaml

If these files are unavailable, the install likely copied a skill without the sibling references/ folder. Tell the user to copy or symlink references/ beside the parent of the copied skills/ folder before continuing.

Workflow

1. Resolve memory config:

• Prefer .naval/config.local.yaml in the current repo.
• If config exists with memory_enabled: false, treat persistence as disabled. Do not search or write under memory_root unless the user explicitly asks for a one-off save path or chooses to run n-setup to enable memory.
• If no config exists, ask whether to continue with no write, write once to a user-provided path, or run n-setup.

1. Extract at most 1-3 learnings from the completed session. Look for:

• a corrected assumption
• a reusable decision rule
• an experiment that should be repeated
• a practice that worked
• a scorecard signal that should guide future choices

1. Discard material that was merely discussed. Do not save private details unless they are necessary and the user approves the privacy level.
2. Draft each learning with:

• title
• type: insight, correction, playbook, pattern, or practice
• tags for future retrieval
• confidence: high, medium, or low
• privacy: public-safe, personal, or private
• why it matters later

1. If persistence remains disabled, return the drafted learnings as unsaved output and stop before duplicate search or writes.
2. Search the approved memory root before writing. Check learnings/, reviews/, decisions/, scorecards/, experiments/, and practices/ for duplicate or superseded entries.
3. Present the draft and duplicate findings. Ask for approval before creating or updating any file.
4. Write approved learnings to <memory_root>/learnings/YYYY-MM-DD-short-title.md using references/memory/templates/learning.md.

Do Not

• Do not auto-save.
• Do not save the full conversation.
• Do not store private reflections in a public repo unless the user explicitly chooses that.
• Do not create five or more learnings from one session; filter harder.

Output

Return:

• Saved files or skipped reason
• Duplicate or supersession handling
• Privacy level used
• Tags that will help future retrieval