The Agent Skills Directory

[SAFE] (SAFE): No malicious patterns or security risks detected across all files. The skill consists of Markdown documentation, JSON test cases, and non-executable code snippets.
[Indirect Prompt Injection] (LOW): The skill is designed to ingest and review user-provided code, which is a surface for indirect prompt injection.
Ingestion points: User prompts and pasted code snippets for review as documented in README.md and evals/evals.json (specifically eval-02-code-review).
Boundary markers (absent): The provided reference files do not explicitly define delimiters for untrusted user code; however, this is expected for reference-only files.
Capability inventory: System architecture design, API endpoint generation, and code review findings. The README.md mentions the use of a code_execution tool in the API configuration section.
Sanitization (absent): No specific sanitization logic is present in the provided reference materials, as the skill provides structural patterns rather than executable sanitization scripts.
[Unverifiable Dependencies] (SAFE): The README.md and CONTRIBUTING.md contain standard CLI instructions (e.g., git clone, cp) for manual user installation. These are not automated or hidden execution patterns.
[Credentials Unsafe] (SAFE): While some files (e.g., references/stacks/java.md, evals/evals.json) contain strings like "password123", these are clearly documented as test data or examples and do not constitute credential exposure.

senior-backend-dev