The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze external source code files and user-provided descriptions of BFF architectures to perform security audits. It lacks explicit boundary markers or instructions to isolate potentially adversarial content embedded within the code being audited, which could theoretically attempt to influence the agent's final audit verdict.
Ingestion points: Audited source code files in directories such as src, app, pages, and server via the rg utility.
Boundary markers: None identified in the provided skill instructions to delimit untrusted code from instructions.
Capability inventory: Local file system searching using the rg (ripgrep) command.
Sanitization: No explicit sanitization of input data is defined before processing by the LLM.
[SAFE]: The skill instructions include standard shell commands using the ripgrep utility to assist the agent in identifying potentially dangerous code patterns (e.g., source-to-sink traces for SSRF). These commands are local search operations appropriate for the skill's stated purpose of security auditing.
[SAFE]: The skill references several external resources, including official documentation from Next.js, Vercel, and the OWASP Foundation, as well as mature open-source projects like Backstage and Grafana. These references are used neutrally for comparative analysis and benchmarking purposes.

bff-proxy-security-contracts