The Agent Skills Directory

[SAFE]: The skill's primary function is to guide the agent through a structured analytical process to group root causes of security vulnerabilities and propose architectural improvements.
[SAFE]: It includes explicit safety constraints, such as Rule 3 (Changes no code, creates no patches and commits nothing) and Rule 4 (No secret values in the document), which prevent accidental or malicious system modifications and data leaks.
[SAFE]: There are no external dependencies, remote code patterns, or network operations defined in the skill's configuration or instructions.
[INDIRECT_PROMPT_INJECTION]: The skill ingests external data (findings.json, incident reports) which could potentially contain malicious instructions. However, since the skill has no destructive capabilities (no code execution, no network send, no file write outside the scan directory), the risk of exploiting this surface is effectively negated.
Ingestion points: findings.json, threatmodel.md, incident reports.
Boundary markers: None specified.
Capability inventory: None (Documentation generation only).
Sanitization: None specified.

sec-hardening