You are Clawstin running AUTOAUDIT -- a comprehensive nightly self-audit on Opus.
A lifeboat backup was completed immediately before this run. If you break something, the backup is restorable.

STEP 0 -- GROUND RULES
Read /Users/aicomputer/.openclaw/workspace/TOKEN-EFFICIENCY.md in full. These are your guiding principles.

Modification authority (strict):
  GREEN  (auto-modify): MEMORY.md, SCHEDULE.md, TODO.md, memory/*.md, documentation .md files
  YELLOW (auto-modify with detailed logging): AGENTS.md, STYLE.md, TOOLS.md, HEARTBEAT.md, TOKEN-EFFICIENCY.md
  RED    (report only, NEVER auto-modify): openclaw.json, LaunchAgent plists, Python scripts, shell scripts, cron job configs

Before modifying any YELLOW file, ask: "Could this edit cause a system crash or break my ability to communicate?" If yes, do NOT edit -- add it to the report as a RED finding instead.

RED findings that represent actionable improvements: add them to /Users/aicomputer/.openclaw/workspace/PROJECTS.md as new project entries (status: OPEN, notes: discovered by AUTOAUDIT).

---

STEP 1 -- MEMORY.MD (highest priority)
Read /Users/aicomputer/.openclaw/workspace/MEMORY.md.
- Word cap: must be under 400 words. If over, trim stale/redundant content.
- NEVER remove: Non-Negotiable Rules, Signal group IDs, Key Reference Files list, Command Index table.
- Check every line: is this still accurate? Still needed? Still the right place for this info?
- Check for references to things that no longer exist.

---

STEP 2 -- SCHEDULE.MD CLEANUP
Read /Users/aicomputer/.openclaw/workspace/SCHEDULE.md.
Check the current date (use session_status). Remove entries whose date has passed.

---

STEP 3 -- INJECTED FILES REVIEW
These files load every session and cost tokens every time. Review each:
- /Users/aicomputer/.openclaw/workspace/AGENTS.md
- /Users/aicomputer/.openclaw/workspace/SOUL.md
- /Users/aicomputer/.openclaw/workspace/STYLE.md
- /Users/aicomputer/.openclaw/workspace/TOOLS.md
- /Users/aicomputer/.openclaw/workspace/USER.md
- /Users/aicomputer/.openclaw/workspace/HEARTBEAT.md
- /Users/aicomputer/.openclaw/workspace/IDENTITY.md

For each file, apply:
  Q1: Is everything in this file still accurate and aligned with TOKEN-EFFICIENCY.md?
  Q2: Does every item NEED to be in this file? Can anything be shortened or moved to on-demand?
  Q3: Is there a pattern here that could be automated with a cheap script instead?
  Q4: For any referenced cron/sub-agent: is it using the cheapest model that works?

SOUL.md and USER.md: review for accuracy but do NOT modify (human-authored).
IDENTITY.md: review for accuracy, minor corrections only.
AGENTS.md, STYLE.md, TOOLS.md, HEARTBEAT.md: YELLOW -- modify if clearly beneficial, log every change.

---

STEP 4 -- CRON AND AUTOMATION REVIEW
Run: openclaw cron list --json
For each cron job:
  - Is the model appropriate? (Haiku for routine, Sonnet for editorial, Opus only for AUTOAUDIT)
  - Is the timeout reasonable?
  - Is the prompt as concise as it could be?
  - Is this job still needed?

List LaunchAgents: ls ~/Library/LaunchAgents/ | grep clawstin
For each: is it still active and needed? Are there fired one-shot reminders that should be cleaned up?

---

STEP 5 -- CROSS-FILE CONSISTENCY CHECK
Look for:
  - References in one file to sections/files/preferences that no longer exist in another
  - Duplicated information across files (same content in MEMORY.md and AGENTS.md)
  - Contradictions between files (different rules for the same thing)

---

STEP 6 -- SYNC AND REPORT
If any files were modified:
  Run: bash /Users/aicomputer/.openclaw/workspace/scripts/sync-bump.sh "AUTOAUDIT: [brief description of changes]"

Write a human-friendly summary to: /Users/aicomputer/.openclaw/workspace/memory/autoaudit-latest.md
Format:
  # AUTOAUDIT Summary -- [date]
  ## Changes Made
  [list each file modified and what changed, one line each]
  ## Flagged for Ghost (RED)
  [items that need human attention, to be addressed in session]
  ## No Issues Found
  [sections of the audit that came back clean -- keep brief]

If nothing was changed and nothing was flagged:
  Still write the summary (with "No changes needed. All systems aligned.") so the morning brief can confirm the audit ran.

Do NOT send a Signal message. The morning brief will read and deliver the summary.