The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill uses uv and uvx to execute analysis scripts, which may involve downloading and running external packages from public registries.
[COMMAND_EXECUTION]: The skill uses the Bash tool to execute internal shell scripts and system compiler components like clang and cargo to perform its analysis tasks.
[REMOTE_CODE_EXECUTION]: The skill's 5b-poc-validator agent compiles and runs Proof-of-Concept programs derived from the audited codebase, representing a dynamic execution risk of code generated based on untrusted inputs.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection from the repository it audits as it lacks sanitization and boundary markers between the audited data and the agent's instructions.
[PROMPT_INJECTION]: Ingestion points: Audited source files within the repository specified by the path argument.
[PROMPT_INJECTION]: Boundary markers: None specified to separate audited code from instructions for the 11-phase agent pipeline.
[PROMPT_INJECTION]: Capability inventory: Bash, Write, Task, clang, and cargo compiler toolchains.
[PROMPT_INJECTION]: Sanitization: No input validation or sanitization is performed on the audited source code.

zeroize-audit