The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from existing pull requests which could theoretically contain malicious instructions.
Ingestion points: SKILL.md (Phases 0 and 5b) reads pull request titles and bodies using the gh pr view command.
Boundary markers: No specific delimiters or instructions to ignore embedded commands are present when the agent reads the existing PR description.
Capability inventory: The skill uses git, gh, and python3 to modify the repository and update pull requests.
Sanitization: The skill does not explicitly describe sanitization for fetched PR content before re-validating it against the current branch state.
[REMOTE_CODE_EXECUTION]: The skill executes repository-provided scripts and a local linter.
Evidence: Phase 3 of the workflow involves running the repository's own formatters, linters, and test suites.
Context: This is a standard requirement for a development-focused AI agent and is performed using the repository's configured task runners.
[COMMAND_EXECUTION]: The skill instructs the agent to perform Git operations, including pushing to remote branches and managing detached references for image hosting.
Evidence: references/screenshots.md provides shell commands for creating git blobs and pushing to refs/attachments/.
Context: These are standard operations for managing pull request assets within a Git-based development workflow.

submit