The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill parses untrusted repository files to discover test and linting commands to execute. This creates a risk where a malicious repository could contain harmful commands in its configuration files that the agent would then execute.
Ingestion points: CONTRIBUTING.md, AGENTS.md, package.json, Makefile, and CI configuration files in the target repository.
Boundary markers: None; the skill instructions tell the agent to find and run these commands without explicit safety delimiters or validation of the command content.
Capability inventory: Shell command execution (via git, thurview, and the repo's own scripts).
Sanitization: None; the skill assumes commands found in the repository are safe to run for testing purposes.
[COMMAND_EXECUTION]: The skill uses user-supplied $ARGUMENTS to construct shell commands such as git switch <branch>, gh pr checkout <n>, and thurview graph impact. If the input is not strictly validated or sanitized by the agent, it could lead to command injection.
[EXTERNAL_DOWNLOADS]: The skill utilizes npx -y thurview to fetch and run the thurview package from the NPM registry if it is not present on the system. It also includes a thurview update command to update the tool from remote sources.

review-fix