The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external data from GitHub Pull Request comments, bot reviews, and CI logs.
Ingestion points: scripts/pr-status.ts fetches PR body, review threads, issue comments, and review summaries via the GitHub API.
Boundary markers: The skill uses a specific attribution line (_[<model-id>] responding on behalf of <PR owner>_) to distinguish its comments and includes explicit instructions to treat bot and comment text as untrusted data.
Capability inventory: The skill allows the agent to execute git push --force-with-lease, gh api (write operations), gh pr edit, and gh run rerun.
Sanitization: The instructions explicitly forbid interpolating external comment text into shell commands, instead mandating the use of mktemp to pass reply bodies as files to the gh CLI, effectively preventing command injection.
[COMMAND_EXECUTION]: The skill relies on local command-line tools for repository and PR management.
Evidence: Extensive use of git for branch management (pull, rebase, fetch, push) and the gh CLI for interacting with GitHub's PR and Action features.
[EXTERNAL_DOWNLOADS]: The skill fetches data from external sources during execution.
Evidence: The pr-status.ts script fetches PR metadata, comments, and review status from GitHub's API via the gh CLI. GitHub is a well-known service and the repository tester-army/e2e belongs to the skill's authoring organization.

babysit