The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill coordinates work based on external task plans and tickets, creating a surface for potential instruction injection through data.
Ingestion points: Plan tasks and tickets supplied in implementation mode (references/implementation-mode.md).
Boundary markers: Work is scoped within "handoff briefs" and isolated Git worktrees for each worker to provide execution boundaries.
Capability inventory: Subagents have the capability to edit files, run validation tools, and create commits; the lead possesses integration authority.
Sanitization: The workflow mandates a "fresh independent read-only reviewer" for final joined branches and requires lead validation of each task commit before integration to mitigate untrusted inputs.
[COMMAND_EXECUTION]: The skill instructs the agent to perform Git repository operations to manage worker environments.
Evidence: Instructions for using git worktree add, git worktree list --porcelain, and git worktree remove to manage isolated task environments (references/implementation-mode.md).
[EXTERNAL_DOWNLOADS]: The documentation references an external utility tool as a prerequisite.
Reference: Mentions Matt Pocock's tdd skill hosted on GitHub.
Context: The skill explicitly states that this is a separately installed prerequisite and mandates that dependencies must never be installed implicitly, requiring explicit user agreement before use.

implement-with-subagents