Agent Bootstrap Hardening

Audit and harden an OpenClaw workspace setup (AGENTS/SOUL/USER/HEARTBEAT/memory) with concise, enforceable rules, safety boundaries, and low-bloat instruction design. Use when users ask to optimize, clean up, or harden their agent brain files.

Use this skill when a user wants to improve core workspace files for better reliability, speed, and safety.

Scope

Target files (if present):

• AGENTS.md
• SOUL.md
• USER.md
• HEARTBEAT.md
• MEMORY.md
• memory/YYYY-MM-DD.md
• optional helper docs: RESOLVER.md, QUALITY.md

Workflow

1. Inventory current files and obvious gaps.
2. Score each file quickly with the rubric in references/rubric.md.
3. Harden content with minimal edits:
   • keep instructions concise and operational
   • tighten safety boundaries and approval rules
   • reduce duplicated or theory-heavy text
   • preserve user tone/persona intent
4. Add missing primitives when useful:
   • RESOLVER.md for execution routing
   • QUALITY.md for lightweight pre-send checks
5. Return summary:
   • what changed
   • why it improved behavior
   • remaining risks/todos

Editing rules

• Prefer small, targeted diffs over wholesale rewrites.
• Avoid introducing process that cannot be followed consistently.
• Keep AGENTS and SOUL readable in one pass.
• Preserve explicit user preferences and platform-specific norms.

Guardrails

• Never weaken privacy or approval boundaries.
• Never add hidden/external exfil behavior.
• Do not claim setup is “secure” without caveats.
• If uncertain about impact, ask before applying broad policy changes.

Output contract

Provide:

• Hardening summary (3–8 bullets)
• Files changed list
• Next recommended step (single highest-leverage follow-up)