The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data including spec.md, files within the tasks/ directory, and project implementation/test files.
Ingestion points: docs/<feature>/spec.md and tasks/*.md are read in their entirety in the Preflight and Discovery steps. Project source code and git history are inspected during Discovery and Review.
Boundary markers: There are no explicit instructions to the subagents to ignore or isolate instructions that might be embedded within the project files or specifications.
Capability inventory: The subagents have the ability to write files (Step 3/5), execute other skills (unity-implement, unity-code-review), and inspect repository state.
Sanitization: The skill lacks sanitization; in Step 5, review findings are passed 'verbatim' to the worker subagent, potentially propagating malicious content found during the review phase into the implementation phase.
[COMMAND_EXECUTION]: The skill executes git rev-parse HEAD in Step 1 to establish a fixed point for the repository state. This is a standard and benign operation for a development-focused skill.

unity-orchestrate