The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external data (GitHub PR comments) and instructs the agent to implement suggested changes, creating a vulnerability to malicious instructions embedded in review feedback.
Ingestion points: The scripts/pr_review.py script fetches review comments from the GitHub API and writes them to local markdown files.
Boundary markers: The instructions do not employ explicit delimiters or system-level warnings to distinguish between the skill's instructions and the content of the review comments.
Capability inventory: The skill provides the agent with capabilities to modify source code, commit changes via git, and execute testing/linting commands via pnpm and uv.
Sanitization: scripts/pr_review.py includes a cleanup_html_text function that performs basic HTML tag removal, but does not sanitize potential natural language prompt injections.
[COMMAND_EXECUTION]: The skill utilizes shell commands for repository management and workflow execution.
The script scripts/pr_review.py executes git config --get remote.origin.url using subprocess.run to identify the target repository.
SKILL.md directs the agent to execute project-specific commands like pnpm run lint, pnpm run typecheck, and pnpm run test, as well as git and gh CLI commands.

fix-coderabbit-review