The Agent Skills Directory

[SAFE]: The skill is strictly instructional and provides a framework for performing architectural audits without introducing external code, scripts, or dependencies.\n- [SAFE]: No hardcoded credentials, sensitive file access (like SSH or AWS keys), or data exfiltration attempts were found. The search patterns target standard application configuration files for topology detection.\n- [SAFE]: No base64, homoglyph, or other obfuscation techniques are present in the skill body or references.\n- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted code from the repository, creating a surface for indirect prompt injection. However, this is inherent to its diagnostic function and is mitigated by its read-only nature.\n
Ingestion points: Repository files searched using rg and fd commands specified in SKILL.md and references/patterns.md.\n
Boundary markers: Explicit instructions require the agent to cite file:line evidence and follow a "fit before presence" logic to justify findings.\n
Capability inventory: The skill is explicitly read-only and lacks capabilities for file writing, network exfiltration, or command execution beyond diagnostic searches.\n
Sanitization: No specific sanitization of file content is defined, though the diagnostic context limits the impact of potential injection.

audit-backend-architecture