The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill analyzes arbitrary source code comments from external repositories, which serves as a vector for indirect prompt injection. Maliciously crafted comments could attempt to influence the agent's classification or fixing logic.
Ingestion points: The skill reads line, block, and documentation comments from JavaScript, TypeScript, and Go source files discovered in the user-provided paths.
Boundary markers: No specific delimiters or "ignore instructions" warnings are defined for the comment content being analyzed.
Capability inventory: The agent has the ability to read files, write fixes to the filesystem, and execute system commands (git, rg, linters, build tools).
Sanitization: There is no evidence of sanitization or filtering of the comment text before it is processed by the analysis subagents.
Safeguards: The instructions include significant safeguards, such as prohibiting the agent from changing executable code merely to satisfy a comment and requiring all writes to be handled in the main context rather than by subagents.
[COMMAND_EXECUTION]: The skill executes various development tools to discover files and verify fixes. This includes repository-aware discovery tools (git ls-files, rg) and project-specific validation tools (formatter, lint, typecheck, test, build). While these are standard for code maintenance, they involve executing binaries within the project context.

stale-comments-audit