The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data, including code files and reviewer comments, which creates a surface for indirect prompt injection.
Ingestion points: Ingests reviews from files, PR comments, or chat messages and reads project source code using Grep, Glob, and Read tools.
Boundary markers: The skill instructs the agent to restate every finding in a normalized one-line format and label reviewers, which helps maintain context and prevents external instructions from blending into the agent's logic.
Capability inventory: The skill is restricted to read-only tools (Grep, Glob, Read). It does not have access to tools for writing files, network access, or shell execution.
Sanitization: No specific sanitization of code comments is implemented, but the risk is mitigated by the lack of any execution or writing capabilities.
[EXTERNAL_DOWNLOADS]: The README provides standard installation instructions that involve fetching the skill from the vendor's GitHub repository.
Evidence: The installation uses npx skills add https://github.com/pilniczek/dev-skills to vendor the skill.
Context: This is a documented installation pattern for the environment and references the official repository associated with the skill's author.

merge-reviews