Workspace Hygiene — Auto-Clean for AI Agents

Automatically enforces workspace file size limits, archives stale or bloated files, and organizes root contents to maintain efficient AI agent context.

Security Scan

The skill largely implements workspace cleanup as described, but there are mismatches between its prose and the actual script and a few potentially destructive behaviors (truncation/deletion) that warrant caution before enabling it.

Purpose & Capability

Name/description match the provided script: it trims large markdown boot files, archives stale files, and moves media/package files out of the workspace root. The requested resources (none) are proportional to this purpose.

Instruction Scope

SKILL.md contains rules and triggers that partly do not match the script: SKILL.md promises 'Never delete, always archive', yet the script deletes empty files; SKILL.md mandates '.secrets' be moved to talonforge/.secrets/, but the script does not touch .secrets; SKILL.md mentions auto-trigger behavior (after sessions that create 10+ files) but the script has no event-driven mechanism. These discrepancies could lead to unexpected outcomes.

Install Mechanism

There is no install spec (instruction-only), which is low-risk. However, the skill includes a runnable shell script in the package that will be executed by the agent if invoked — the script operates on the filesystem but does not fetch external code or contact networks.

Credentials

No environment variables or external credentials are requested. The script operates on a single workspace path (/root/.openclaw/workspace) which is consistent with a workspace hygiene tool, but this absolute path may behave differently depending on the agent runtime and privileges.

Persistence & Privilege

The skill is not 'always: true' and is user-invocable; it does not request persistent elevated privileges or modify other skills' configs. Autonomous invocation is allowed by default but not a unique risk here.

What to consider before installing

This skill appears to implement a workspace cleaner and archives many files locally (no network exfiltration). Before installing: (1) back up your workspace (AGENTS.md, MEMORY.md, STATE.md, etc.) because the script can truncate/trim files and will delete empty files; (2) review and test the included workspace-hygiene.sh in a sandbox or copy it to a non-production workspace to verify behavior and portability (stat/date flags may be platform-dependent); (3) confirm the hard-coded WORKSPACE path (/root/.openclaw/workspace) matches where your agent stores data or modify it to your environment; (4) note mismatches between SKILL.md and the script (the script does delete empty files and does not move .secrets), so ask the author to reconcile or adjust the script if you require stricter 'never delete' guarantees; and (5) if sensitive secrets or unique files reside under the workspace root, ensure they are excluded or moved before enabling automatic runs.

Workspace Hygiene Skill

Auto-maintains workspace cleanliness. Prevents context bloat that burns tokens on every session boot.

Triggers

• Cron: runs daily at 06:00 Europe/Berlin
• Manual: when Zinou says "clean up", "audit workspace", "hygiene check"
• Auto: after any session that creates 10+ files

Rules

1. BOOT FILE SIZE LIMITS (hard limits, enforce or trim)

• AGENTS.md: max 50 lines
• MEMORY.md: max 80 lines
• STATE.md: max 60 lines
• SOUL.md: no limit (identity, changes rarely)
• USER.md: max 40 lines
• IDENTITY.md: max 30 lines
• TOOLS.md: no limit (reference, not loaded often)
• HEARTBEAT.md: max 20 lines

If any file exceeds its limit:

1. Identify bloat (action logs, duplicate info, stale references)
2. Move old content to memory/archive/YYYY-MM-DD-<topic>.md
3. Leave a 1-line pointer: See archive/YYYY-MM-DD-<topic>.md
4. Log the trim in today's daily note

2. ROOT WORKSPACE RULES

• NO image/video files in workspace root → archive/frames/ or archive/media/
• NO package.json/package-lock.json in root (belongs in subprojects)
• NO temp scripts in root → /tmp/ for those
• NO .secrets in root → always in talonforge/.secrets/
• Only .md files and directories allowed in workspace root

3. MEMORY/ RULES

• Only keep files actively needed this week
• Daily notes older than 3 days → archive (except today's)
• Stale files (not updated in 7+ days, not referenced by STATE.md) → archive
• Fragment daily notes (multiple per day) → consolidate into one
• claude-task.md → archive when >7 days stale, keep latest 500 lines
• Never delete, always archive to memory/archive/

4. AUTO-ARCHIVE PATTERNS

After any session:

• If temp files created in /tmp/ → leave them (auto-cleaned)
• If files created in workspace root → move to proper location
• If memory/ grows past 10 active files → audit for archiving

5. BLOAT SIGNALS (check for these)

• MEMORY.md action log >5 entries → consolidate to archive
• Same info in 2+ files → keep one, archive the other
• Files with "TODO" or "DRAFT" older than 3 days → flag or archive
• Empty files (<10 bytes) → delete
• Duplicate daily notes → merge

6. ARCHIVE STRUCTURE

memory/archive/
  action-log-day1-3.md
  day3-fragments/
  frames/
  media/
  <date>-<topic>.md

7. REPORT FORMAT

After each run, log to today's daily notes:

## Workspace Hygiene (HH:MM)
- Trimmed: X files, Y lines removed
- Archived: Z files
- Boot load: N lines (was M lines)
- Status: CLEAN / NEEDS ATTENTION

Cron Setup

Run: 0 6 * * * (daily at 06:00 Europe/Berlin) This is a lightweight check — most days it does nothing. Only acts when bloat exists.