Security audit
Security checks for vulnerabilities and agentic risk
This is a markdown-only autonomy framework with broad routing language, but it discloses human-approval boundaries and contains no code, persistence, credential handling, or hidden execution path.
Install only if you want a broad planning/autonomy prompt. Keep normal agent approval controls enabled for production, destructive, security-sensitive, privacy-impacting, financial, or long-running work, and treat its learning/self-correction language as guidance rather than a real memory or automation implementation.
The invocation description is extremely broad, covering general planning, learning, self-correction, task receipt, and autonomous decisions. In a skill-routing environment, such breadth can trigger this high-autonomy skill in situations where a narrower, safer skill should be used, increasing the chance of unintended autonomous action on sensitive tasks.
The manifest description states the skill is capable of adaptive learning and self-correction with minimal supervision. However, the provided skill file is purely instructional markdown containing process descriptions, example logs, and checklists; it does not implement persistence, pattern learning, or any executable mechanism that would allow the agent to actually learn from outcomes.
The description represents the skill as capable of self-correction and autonomous decision-making. In the file, self-correction appears only as written procedures, checklists, and templates for analysis; there is no code or executable logic that detects failures, selects remediations, or applies corrections automatically.
The skill advertises minimal-supervision autonomy up front but does not lead with an equally prominent warning about the risks and limits of autonomous action. That omission can cause users or agent frameworks to overtrust the skill before reaching later safety caveats, especially in systems that primarily consume top-of-file metadata.
The file presents conflicting authority boundaries: the manifest advertises autonomous decisions with no human input, while the body says human input is required for critical, high-risk, and security-sensitive actions. This inconsistency can cause downstream agents or orchestrators to grant broader autonomy than intended, weakening safety controls around sensitive operations.
Skill enables autonomous high-impact decisions without human-in-the-loop verification. Critical operations (destructive commands, financial transactions, data deletion) should require explicit user confirmation.
✅ Document decisions and rationale
### Never:
❌ Take irreversible action without confirmation
❌ Access unauthorized resources
❌ Hide errors or issues
❌ Exceed granted permissions
No suspicious patterns detected.