jarvis-update

Jarvis Brain Maintenance Sweep

Full update of all brain files so the next session inherits complete context.

Environment Detection

Before starting, detect your runtime:

• Claude Code (CLI): Bash tool available, full filesystem, MCP servers loaded → execute ALL steps
• Claude Cowork (web/desktop): Limited shell, no local-rag MCP, no jarvis-brain API → execute Steps 1-3 (file updates only), SKIP Steps 3.5/3.6 (RAG sync + fact-inbox), note skipped steps in the summary banner

If you cannot run python3 --version via Bash, you are in Cowork — skip infrastructure steps gracefully.

Process

Step 1: Load All Brain Files (in parallel)

Read from workspace root (use CLAUDE.md canonical root on Mac; in Cowork use mounted workspace at the path shown in your environment):

1. Jarvis/status.json — Machine-readable system status
2. Jarvis/MOTD.md — Session startup banner
3. Jarvis/project-status.md — Active project tracker
4. Jarvis/recommendations.md — Prioritized next actions
5. Jarvis/needs.md — Self-aware limitation tracking
6. Jarvis/growth-tracker.md — Capability evolution log
7. memory/history/recent-activity.md — Rolling activity log (last 14 days)
8. Jarvis/audit/session-log.md — Rolling session log (last 10 sessions)

Step 2: Assess What Changed This Session

Review the conversation and identify:

• What tasks were completed?
• Were any deliverables created or updated?
• Did data freshness change (new data pulls)?
• Were any new needs or blockers identified?
• Did priorities shift based on work done?
• Any lessons learned or mistakes to capture?
• Were new capabilities added (tools, commands, features)?
• Were any projects advanced or completed?

Step 3: Update Each File

Update ALL files that need changes. Typical updates:

Jarvis/status.json

• Update last_updated timestamp
• Update updated_by to current model
• Update recent_sessions array (add new entry at top, keep last 6)
• Update deliverables.completed if new deliverables produced
• Update mcp_servers if server status changed
• Update web_console if new features/endpoints/tabs added
• Update instance data if refreshes occurred

Jarvis/MOTD.md

• Update RECENT ACTIVITY section with current session work
• Update SYSTEM STATUS if instances/MCP status changed
• Update DELIVERABLES PRODUCED if new ones created
• Update CAPABILITIES if new tools/features added
• Update QUICK COMMANDS if new commands created
• Bump version number if significant capability change

Jarvis/project-status.md

• Update milestone statuses (Planned → Complete with date)
• Update project percentages
• Update "Next Step" and "Blockers" fields
• Add new projects if started
• Move completed projects to Completed section

Jarvis/recommendations.md

• Move completed recommendations to "Completed" section
• Add new recommendations discovered during session
• Re-rank remaining items based on current priorities
• Update Tips & Insights with new learnings
• Update Last Updated date

Jarvis/needs.md

• Mark resolved needs as completed
• Add any new needs discovered
• Update Last Updated date

Jarvis/growth-tracker.md

• Add new capability entries with dates
• Log any infrastructure changes

memory/history/recent-activity.md

• Add new session header if needed (### Session N (description))
• Add bullet points for each accomplishment
• Keep entries concise (one line per item, dash prefixed)

Jarvis/audit/session-log.md

• Add new session entry with: date, duration estimate, key actions, files modified
• Keep last 10 sessions only (archive older entries monthly)

Step 3.5: Sync RAG Vector Index

Run the RAG sync scanner to detect changes made during this session:

1. Run: python3 scripts/jarvis/rag_sync.py --scan
2. Review the output actions (add/update/remove)
3. For each file in add or update: call mcp__local-rag__ingest_file with the file path
4. For each file in remove: call mcp__local-rag__delete_file with the file path
5. Run: python3 scripts/jarvis/rag_sync.py --update
6. If no changes detected, skip and note "RAG index: no sync needed"

Step 3.7: Improvement Cadence Check (Wave 52)

Check if the weekly improvement cadence is overdue:

1. Read Jarvis/status.json → improvement_cadence.last_weekly_cadence
2. If null or >7 days old: add "⚡ CADENCE OVERDUE" to the summary banner
3. Read config/wave-priorities.json — surface the top-scored planned wave in the banner
4. If cadence is overdue, recommend running cadence steps before the next work session
5. Check improvement_cadence.last_competitive_scan — if >7 days, flag scan as overdue too

Step 3.6: Process Fact Inbox

Check sessions/fact-inbox.jsonl for pending facts from other agents (Codex, Antigravity):

1. Read sessions/fact-inbox.jsonl — if empty or missing, skip and note "fact-inbox: empty"
2. For each JSON line, call POST http://localhost:8000/api/memory/facts with content, category, and source_agent
   • Or use memory_engine.add_fact(content, category) if the API is not reachable
3. After processing all lines, clear the file: write an empty file back to sessions/fact-inbox.jsonl
4. Log: "fact-inbox: N facts ingested" in the Step 4 summary banner

Step 4: Confirm

Present a summary banner:

BRAIN MAINTENANCE COMPLETE | [date] — Session [N]
══════════════════════════════════════════════════
  status.json:        ✅ [describe changes]
  MOTD.md:            ✅ [describe changes]
  project-status.md:  ✅ [describe changes]
  recommendations.md: ✅ [describe changes]
  needs.md:           ✅ [describe changes]
  growth-tracker.md:  ✅ [describe changes]
  recent-activity.md: ✅ [describe changes]
  session-log.md:     ✅ [describe changes]
  rag-sync:           ✅ [N files synced / no sync needed]
  cadence:            ✅ [current / ⚡ OVERDUE (N days)]
  next-wave:          [top-scored planned wave from wave-priorities.json]
══════════════════════════════════════════════════
  All brain files synced. Next session picks up here.

If running in Cowork (no Bash/MCP), use this reduced banner:

BRAIN MAINTENANCE (COWORK) | [date] — Session [N]
══════════════════════════════════════════════════
  status.json:        ✅ [describe changes]
  MOTD.md:            ✅ [describe changes]
  project-status.md:  ✅ [describe changes]
  recommendations.md: ✅ [describe changes]
  needs.md:           ✅ [describe changes]
  growth-tracker.md:  ✅ [describe changes]
  recent-activity.md: ✅ [describe changes]
  session-log.md:     ✅ [describe changes]
  rag-sync:           ⏭ skipped (Cowork — no local-rag MCP)
  fact-inbox:         ⏭ skipped (Cowork — no jarvis-brain API)
  cadence:            ✅ [current / ⚡ OVERDUE (N days)]
══════════════════════════════════════════════════
  Brain files synced. RAG + fact-inbox deferred to next Code session.

Rules

• Never delete history entries, only append
• Keep recent-activity.md entries to one line each
• Maintain existing file formats exactly
• Update ALL files — this is a comprehensive sweep, not a partial update
• Always increment session number from the most recent entry
• If a file doesn't need changes, say "no changes needed" in the summary
• If unsure whether something changed, include it (err on the side of completeness)