The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructions reference an external repository mattpocock/skills for the /tdd functionality. This constitutes a dependency on third-party code from a source outside the verified vendor scope.
[COMMAND_EXECUTION]: The skill performs several file system and version control operations as part of its workflow, including:
Executing git rev-parse HEAD and git log to track state.
Creating symbolic links (symlinking CLAUDE.md to AGENTS.md).
Deleting ledger files (progress.txt and plans/*.md) during the 'purge' phase.
[PROMPT_INJECTION]: The skill possesses an indirect prompt injection surface due to its reliance on untrusted external data:
Ingestion points: The agent reads and trusts instructions from progress.txt, CONSTITUTION.md, and user-provided plan files in plans/*.md (SKILL.md).
Boundary markers: Absent. The skill explicitly instructs the agent to "trust it [progress.txt] ... over your own memory."
Capability inventory: The controller agent has the capability to spawn subagents (Implementers and Reviewers) and provide them with content extracted from these files. It also manages file writes to AGENTS.md and CONTEXT.md.
Sanitization: Absent. The instructions mandate passing findings and task sentences verbatim to subagents during the dispatch and fix loops.

work-implement