The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted content from GitHub PR comments and reviews, which is a surface for indirect prompt injection. The skill includes robust mitigations to handle this safely.
Ingestion points: github_cli.dart fetches PR comments, reviews, and CI logs using the GitHub API (gh api).
Boundary markers: The instructions include a 'Critical Mindset' section that explicitly warns the agent that 'Reviewer Feedback is NOT Gospel' and mandates an empirical verification gate (dart analyze) before accepting any reviewer claims.
Capability inventory: The skill can execute shell commands via Process.run (to call git and gh), write to the filesystem, and perform network operations via the GitHub CLI.
Sanitization: The skill relies on strict workflow instructions, empirical verification of claims, and user approval gates to sanitize the impact of potentially malicious instructions embedded in PR comments.
[COMMAND_EXECUTION]: The skill uses local subprocesses to interact with repository management tools.
Evidence: lib/pr_context.dart defines a runCommand function that executes git and gh binaries using Process.run. These operations are limited to the target repository and the GitHub API, consistent with the intended purpose of the skill.

github-pr-triage