The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill creates an attack surface for indirect prompt injection by design, as subagents process external data from GitHub to perform automated code changes.
Ingestion points: Subagents read the issue body and PR comments fetched via gh api (specifically gh api repos/.../pulls/.../comments).
Boundary markers: The instructions mandate "self-contained prompts" for subagents, including "explicit out-of-scope boundaries" to prevent unintended context inheritance.
Capability inventory: The workflow utilizes git for code modification (commit, push) and gh for administrative PR actions (review, merge).
Sanitization: The skill explicitly requires a bail-out contract and a mandatory redaction policy for credentials and local file paths in all outputs.
[COMMAND_EXECUTION]: The skill uses standard CLI tools (git, gh) to manage the software development lifecycle. These are executed to perform repository operations, worktree management, and GitHub API interactions. It also references project-specific tooling such as ai-agent-pr-metadata for metadata management and disposition gates.

subagent-pipeline