The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the agent to install oxlint-tsgolint as a devDependency. This package does not appear to be a standard or well-known component of the TypeScript or Oxc ecosystems, posing a risk of dependency confusion or the installation of untrusted code from an unknown source.
[INDIRECT_PROMPT_INJECTION]: The skill implements a comment-linting system that requires the agent to read and process the content of code comments across the repository. This creates a surface for indirect prompt injection, as malicious instructions could be embedded in these comments to influence the agent's behavior during the 'fixing' phase.
Ingestion points: The lint-comments.mjs script reads all source files (.ts, .js, etc.) in the repository to extract comments for analysis.
Capability inventory: The agent has permissions to write files, modify project configurations (package.json, AGENTS.md), and execute shell commands via the project's package manager.
Sanitization: The script reports the raw content of disallowed comments to the agent without sanitization or filtering of potential instructions.
Boundary markers: The script outputs the path, line number, and raw comment text to stderr. While this provides structure, it lacks explicit boundary markers or instructions for the agent to ignore command-like text within the comments.
[DYNAMIC_EXECUTION]: The skill copies a custom JavaScript template (assets/lint-comments.mjs) into the target repository's scripts/ directory and configures the package manager to execute this script using Node.js. While the script performs legitimate parsing, this pattern of deploying and executing local scripts is a form of dynamic code execution.
[COMMAND_EXECUTION]: The skill executes several shell commands to manage the quality stack, including package manager installations (npm, pnpm, or bun), Git file listing (git ls-files), and running the newly configured check commands.

setup-code-quality