| name | wiki-ingest |
| description | Ingest current session learnings into the Brain-1 wiki at ~/2nd Brain/2nd Brain/Wiki/. Updates existing pages in place, creates new pages for genuinely new topics, appends to log.md. Run at the end of any significant working session.
|
| allowed-tools | Read, Edit, Write, Bash |
Wiki Ingest — Session Learnings
Write this session's durable knowledge into the Brain-1 wiki so Margot has it next time.
Step 1: Orient
Read the index and identify which pages are relevant to what changed this session:
cat ~/2nd\ Brain/2nd\ Brain/Wiki/index.md
Step 2: Identify affected pages
For each significant change or decision from this session, match it to an existing page.
Cap at 10–15 pages max. Don't update pages that don't need it.
Common mappings:
- Hermes jobs / cron changes →
hermes-agent.md
- Swarm architecture changes →
pi-ceo-architecture.md
- Wave progress →
wave-roadmap.md
- New business facts → the relevant business page
- New clients or deals →
businesses-overview.md + client page
- System priorities →
operational-priorities-q2-2026.md
Step 3: Update pages in place
For each affected page:
- Read the current page
- Edit ONLY the sections that changed — don't rewrite whole pages
- Update the frontmatter
updated: date to today
- Use
[[double-bracket]] links for cross-references
Format for frontmatter:
---
type: wiki
updated: YYYY-MM-DD
---
Step 4: Create new pages (only if needed)
If a genuinely new topic emerged that doesn't fit any existing page:
- Create
~/2nd Brain/2nd Brain/Wiki/<slug>.md
- Add it to
Wiki/index.md under the right section
Step 5: Append to log.md
YYYY-MM-DD | ingest | page1.md, page2.md | one-line summary of what changed
Step 6: Refresh OKF indexes
Keep the vault navigable (Open Knowledge Format) — regenerate index.md across all
folders so agents read the index first instead of keyword-matching. Idempotent;
preserves hand-authored indexes (those lacking the okf:generated marker).
python3 "$HOME/2nd Brain/2nd Brain/_system/okf-index.py" "$HOME/2nd Brain/2nd Brain"
After regenerating, revert date-only churn so the commit stays surgical — keep only
indexes whose content changed:
git diff --numstat | awk '$1==1 && $2==1 {print substr($0, index($0,$3))}' | tr -d '"' \
| while IFS= read -r f; do git checkout -- "$f"; done
Step 7: Sync to Supabase corpus
SUPABASE_SERVICE_ROLE_KEY=$(grep SUPABASE_UNITE_GROUP_SERVICE_KEY ~/.hermes/.env | cut -d= -f2-) \
python3 ~/Pi-Dev-Ops/scripts/sync_wiki_to_supabase.py
Applied-learning capture (said-it-twice rule)
If you had to re-explain something, hit a repeated failure, or found a workaround for a tool
limitation this session, capture it once — never explain it a third time. Keep each to ONE
bullet under 15 words, no rationale, only things that save a future session. Route it to the
right home: a memory file for agent-behaviour/feedback lessons, the relevant wiki page for
domain knowledge. Adding the fact that matters and dropping the dead one is curation, not
hoarding. See [[feedback-no-status-bloat]].
Rules
- Every sentence carries information — no filler, no commentary
- Sources/ files are immutable — never edit them
- Touch only what this session actually changed
index.md and log.md need updating even if no content pages changed
