The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data by scanning source code comments (TODO, FIXME, HACK) and reading the debt register file. Malicious instructions embedded in these comments could theoretically influence the agent during the scan or report phase.
Ingestion points: Codebase files via Grep/Glob/Read and the docs/tech-debt-register.md file.
Boundary markers: The skill does not define specific delimiters for separating code content from instructions, though it treats findings as data to be tabulated.
Capability inventory: Access to Read, Glob, Grep, and Write tools. The Write tool is specifically directed at updating the technical debt register.
Sanitization: No explicit sanitization of comment content is performed before interpolation into the markdown report.
Mitigating Factor: The skill requires explicit user confirmation before performing any file write operations, providing a human-in-the-loop checkpoint.

tech-debt