The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes external, potentially untrusted data from GitHub PR comments, GitHub Action logs, and local bug report files.
Ingestion points: External data is retrieved via gh pr view (comments), gh run view --log-failed (CI logs), and local files via the --from flag in report-issue.md.
Boundary markers: There are no explicit boundary markers or instructions to the agent to disregard instructions contained within the ingested data.
Capability inventory: The skill possesses significant write capabilities, including git commit, git push --force-with-lease, and various gh api write methods.
Sanitization: No sanitization or validation of the ingested external content is performed before it is processed by the agent.
[REMOTE_CODE_EXECUTION]: The dispatcher in SKILL.md dynamically constructs file paths to load and execute workflows based on the VERB argument (references/workflows/{VERB}.md). This design represents a potential path traversal surface if user input is not properly constrained by the calling environment. Additionally, the ci.md workflow supports an --autofix mode where the agent autonomously generates and pushes code changes to the repository based on failure logs, which involves dynamic modification of executable content.
[COMMAND_EXECUTION]: The skill relies on shell command execution for its core functionality, invoking gh, git, and jq to interact with repository state and the GitHub API.

gh