The Agent Skills Directory

[COMMAND_EXECUTION]: Executes several shell commands for Git and GitHub operations including gh pr view, gh pr checks, gh run view, gh pr comment, and gh pr merge. It also runs the agent-reviews CLI via npx.
[EXTERNAL_DOWNLOADS]: Downloads and executes the agent-reviews package at runtime using npx agent-reviews. While the repository pbakaus/agent-reviews is linked, npx fetches the latest version from the NPM registry, which could be exploited if the package is compromised.
[INDIRECT_PROMPT_INJECTION]: The skill processes external, untrusted data from GitHub pull request comments, review summaries, and CI logs. It classifies these inputs (e.g., 'fix', 'question') and acts upon them. This creates an attack surface where an attacker could embed malicious instructions in a PR comment to influence the agent's behavior, although the skill includes instructions to prefix responses and validate against code.
[PROMPT_INJECTION]: Contains directive language such as 'Never merge from Agent Review or Human Review' and 'Stop and report when...', which are standard task-specific constraints for the AI agent but reflect the instruction-based nature of the skill.

review