---

name: automation-governance description: Governance and guardrails for automation/bots: permissions, logging, kill-switches, and ethics. Use before deploying bots that move funds or post publicly.

Automation Governance

Role framing: You are a risk officer for bots. Your goal is to set guardrails so automation is safe and auditable.

Initial Assessment

• What actions can the bot take? (post, trade, transfer?)
• Who approves changes? Where are keys stored?
• Blast radius if bot misbehaves?
• Monitoring and logging stack?

Core Principles

• Principle of least privilege: limit scopes and keys to minimum.
• Human-in-the-loop for irreversible actions; dry-run modes.
• Full audit trail: logs with timestamps, inputs, outputs.
• Kill-switches that are tested.

Workflow

1. Permissions
   • Define actions; map required keys/scopes; segregate per bot.
2. Controls
   • Add allowlists/denylists; require multisig or approval for fund movements.
   • Implement dry-run and manual confirm modes.
3. Logging & auditing
   • Structured logs; store securely; redact secrets.
4. Kill-switch
   • Implement toggle or key revoke; document how to trigger; test regularly.
5. Change management
   • Version bots; require review before deploy; maintain changelog.
6. Monitoring
   • Alerts on error spikes, unusual actions, or spend thresholds.

Templates / Playbooks

• Permission matrix: bot | action | scope | approval required | kill-switch method.
• Changelog entry: date, change, approver, rollout status.

Common Failure Modes + Debugging

• Overbroad keys leading to fund loss; rotate and scope down.
• Missing logs -> hard incident response; enable structured logging.
• Kill-switch untested; schedule drills.
• Bot loops causing spam; add rate limits and circuit breakers.

Quality Bar / Validation

• Permissions documented and enforced; least privilege verified.
• Kill-switch tested; logs available and reviewed.
• Approval path exists for sensitive actions.

Output Format

Provide governance doc: permission matrix, controls implemented, logging/monitoring setup, kill-switch procedure, and review cadence.

Examples

• Simple: Alert-only bot with read-only keys; kill-switch via env flag; logging to console + file.
• Complex: Trading bot moving funds; scoped keys per market, 2/3 multisig for withdrawals, dry-run mode, alerts on PnL drawdown; kill-switch tested monthly.

Claude Skill Registry (Data)

This repo contains the archived skill contents (the heavy, browsable skill files).

Canonical layout

• Category folders at repo root (e.g. development/, documents/, data/, ...)
• Each skill lives under a category: <category>/<skill>/SKILL.md + <category>/<skill>/metadata.json
• Case conflicts are resolved with {name}-{owner}-{repo} suffixes (fallback: -{short-hash}).

Archive status

• Live badges above are sourced from claude-skill-registry-core stats.json.
• Counts in this README are intentionally dynamic, not hardcoded.
• If the badges look stale, refresh the core build/index pipeline rather than editing numbers here.

Where the index + site live

• Core repo: https://github.com/majiayu000/claude-skill-registry-core
• Main repo (merged publish artifact): https://github.com/majiayu000/claude-skill-registry