⏳ This skill is pending AI review.

Scores will appear once the review pipeline completes.

version unknown

spring-clean

al-matty@al-matty⭐ 0 stars· last commit 7mo ago· 0 open issues

>

Choose how to use this skill

You do not need every option. Choose the path your AI client supports. The stable page stays the same; versioned files are immutable.

1. Native installer

This listing has no registered native installer command. Use the complete package or source fallback below, depending on what your client supports.

Do not guess an installer command or replace an existing version without reviewing the diff.

2. Complete package recommended

Download the ZIP when available. It includes SKILL.md plus the references, security notes and version metadata.

No complete ProSkills package is published for this listing yet.

3. Prompt-only

Copy the prompt above when the agent can read the stable page or when you want to adopt the workflow without installing a skill.

Need only the instruction file?

Download SKILL.md only if your client requires a single file. The complete ZIP is safer for a full installation because it preserves the references and release context.

No path installs or executes anything by itself. Your agent still needs access to the project files. Before updating, compare the installed version and review the diff.

—/10

// RATINGS

⭐GitHub Stars
⭐ 0 on GitHubGitHub ↗

New / niche

🟢ProSkills Score
—
📍

Not yet listed on ClawHub or SkillsMP

// README

OpenClaw Skills

A collection of workspace skills for OpenClaw, the open-source AI agent framework.

Available Skills

SkillDescription
spring-cleanAudit workspace files for token bloat, cross-file conflicts, redundancy, stale content, and misplaced information. Produces a diff-preview for human approval before any writes.

What are workspace files?

In agent frameworks like OpenClaw, Claude Code, or Codex, files like AGENTS.md, TOOLS.md, and USER.md give the LLM persistent instructions, preferences, and domain knowledge. They're injected into every API call as part of the system prompt.

The downside: they make every prompt more expensive and eat into your context window. These skills help you manage that tradeoff.

Installation

Via ClawHub (recommended):

clawhub install al-matty/spring-clean

Manual:

Copy the skill folder into your OpenClaw workspace:

cp -r spring-clean ~/.openclaw/workspace/skills/

The skill will be available on your next session.

Context

These skills were born out of running OpenClaw as a daily-driver personal assistant. Blog posts with more background:

  • Coming soon: workspace file management for AI agents

License

MIT

// HOW IT'S BUILT

// FILES

2 files shipped with this skill. Inline text is the catalog copy. The upstream link is the source repository.

  • spring-clean/SKILL.md6.0 KBupstream ↗
    ---
    name: spring-clean
    description: >
      Workspace file audit: check injected file token budgets, cross-file conflicts,
      redundancy, stale info, misplaced content, and unreferenced non-injected files.
      Produces a diff-preview for human approval before any writes.
      Use when: (1) cron triggers a periodic audit, (2) user says "spring clean" or
      "workspace audit" or "check workspace files", (3) after major config/infra changes
      to verify file consistency.
    ---
    
    # Spring Clean — Workspace File Audit
    
    Audit all workspace files for bloat, conflicts, redundancy, and misplacement.
    Output a diff-preview for human approval. Never write without approval.
    
    ## Token Budgets (injected files)
    
    These files are auto-injected into every turn's context window. Keep them lean.
    
    | File | Target | Hard Max | Purpose |
    |---|---|---|---|
    | AGENTS.md | ~700 tok (~2,800B) | 1,000 tok (4,000B) | Behavioral rules only |
    | SOUL.md | ~450 tok (~1,800B) | 600 tok (2,400B) | Persona and tone |
    | USER.md | ~220 tok (~900B) | 400 tok (1,600B) | Factual profile |
    | TOOLS.md | ~670 tok (~2,700B) | 1,000 tok (4,000B) | Environment-specific tool notes |
    | IDENTITY.md | ~80 tok (~330B) | 150 tok (600B) | Name, emoji, signature |
    | HEARTBEAT.md | ~260 tok (~1,050B) | 400 tok (1,600B) | Heartbeat checklist |
    | MEMORY.md | ~600 tok (~2,400B) | 1,500 tok (6,000B) | Curated long-term memory |
    | **Total** | **~3,000 tok** | **5,000 tok** | — |
    
    MEMORY.md has a higher max because it grows naturally with new projects/lessons.
    
    > **Customizing budgets:** These are sensible defaults for a 200k-token context window.
    > If your workspace uses different files or you have a larger/smaller context window,
    > adjust the targets proportionally. The principle: injected files should use <3% of
    > your total context window.
    
    ## Non-Injected Files (read on demand)
    
    These are NOT auto-injected. They're accessed via `read` or `memory_search`.
    They MUST be referenced from at least one injected file so the agent knows when to consult them.
    
    Common examples:
    
    | File | Purpose | Should be referenced from |
    |---|---|---|
    | memory/*.md | Daily logs, standing instructions | MEMORY.md or AGENTS.md |
    | skills/*/SKILL.md | Skill docs | System prompt (automatic) |
    
    > **Your workspace may have additional non-injected files** (setup guides, goal docs,
    > reference material). Add them to this table if you want the audit to track them.
    
    ## Protected Sections (optional)
    
    Some sections exist for identity, tone, or onboarding reasons and should NOT be
    removed or compressed, even if they seem redundant or "obvious."
    
    To protect specific sections from compaction, add a comment block at the top of
    any workspace file:
    
    ```markdown
    <!-- spring-clean: protect
    - "Section heading or unique phrase"
    - "Another section to preserve"
    -->
    ```
    
    The audit will skip these sections during compaction proposals.
    
    **Default protection (always applied):**
    - **SOUL.md in its entirety** — persona files are never trimmed for token savings.
    
    ## Audit Checklist
    
    Run all checks against ALL workspace files (injected AND non-injected). In order:
    
    ### 1. Size Check
    For each file, measure bytes and estimate tokens (~4 chars/token).
    
    **Injected files:** compare against target and hard max in the budget table above.
    **Non-injected files:** no hard token limit, but shorter is always better. Flag anything
    that seems bloated relative to its purpose.
    
    ### 2. Cross-File Conflict Check
    Same fact stated differently in two files. Check ALL file pairs, not just injected ones.
    Common conflict zones:
    - Config values (settings, ports, paths) across files
    - Behavioral rules appearing in multiple files with different wording
    - Personal info duplicated with different details
    - Tool instructions split across files with contradictions
    
    ### 3. Redundancy Check
    Content appearing in multiple files. Check ALL files. Rules:
    - Each fact has ONE canonical location. Cross-references are fine; duplicated content is not.
    - Decide canonical locations based on the file's purpose (see the budget table).
    
    ### 4. Misplacement Check
    Content in the wrong file based on each file's stated purpose.
    
    ### 5. Staleness Check
    ALL files. Look for: completed tasks listed as active, old version numbers, resolved
    issues, expired reminders, done TODO items, outdated status markers.
    
    ### 6. Reference Check (non-injected files)
    For each non-injected file:
    - Verify it's referenced from at least one injected file
    - If unreferenced: flag as "orphaned — agent won't know to read this"
    
    ### 7. Rule Quality Check
    For each behavioral rule found in any file:
    - **Default behavior?** Model already does this without being told → candidate for removal
    - **Vague?** Would be interpreted differently each time → rewrite or remove
    - **One-off fix?** Reads like it was added after one bad output → consider removing
    - **Verbose?** Can it be said in fewer words → rewrite
    
    ## Output Format
    
    Produce a structured report:
    
    ```
    🧹 Spring Clean Report — YYYY-MM-DD
    
    📏 SIZE CHECK
      AGENTS.md:    XXX tok (target 700, max 1000) ✅/⚠️/🔴
      MEMORY.md:    XXX tok (target 600, max 1500) ✅/⚠️/🔴
      ...
      Total:        XXX tok (target 3000, max 5000)
    
    ⚔️ CONFLICTS (N found)
      - [file1 line X] vs [file2 line Y]: description
    
    🔁 REDUNDANCY (N found)
      - "fact X" in file1 AND file2. Canonical: file1. Remove from: file2.
    
    📦 MISPLACED (N found)
      - "content X" in file1 → should be in file2
    
    🕰️ STALE (N found)
      - "item X" in file1: reason it's stale
    
    🔗 ORPHANED FILES (N found)
      - file1: not referenced from any injected file
    
    ✂️ PROPOSED CHANGES
      ━━ file1 ━━
      - line X: [old] → [new]
      + line Y: [addition]
    
      ━━ file2 ━━
      ...
    
      ━━ No changes: file3, file4
    ```
    
    ## Execution Rules
    
    1. **Never write without approval.** Always present the report and wait.
    2. **Show diffs**, not just descriptions. The human should see exactly what changes.
    3. If no issues found, say so briefly. Don't invent problems.
    4. Track audit history in `memory/spring-clean-state.json`:
       ```json
       {
         "lastAudit": "YYYY-MM-DD",
         "totalTokens": 0,
         "issuesFound": 0,
         "issuesFixed": 0
       }
       ```
    
  • README.mdupstream ↗

    Contents are not inlined in the catalog. Open the upstream file.

Plain-text copy, no login: /skill-bundles/al-matty-openclaw-skills-spring-clean.md

// REPO STATS

0 stars
0 open issues
Last commit: 7mo ago