The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external data that could contain malicious instructions designed to influence the agent.
Ingestion points: Reads from contract.md, story files, and project-specific instruction files (AGENTS.md, CLAUDE.md) within the target repository.
Boundary markers: Employs delimited TASK-LIFECYCLE blocks in contract files to scope execution state.
Capability inventory: Performs file system mutations (patching production and test code), interacts with external domain skills (unity-cli, unity-game-dev), and executes shell-based test batching scripts.
Sanitization: The instructions explicitly forbid interpolating user requests directly into shell commands to prevent injection attacks.
[COMMAND_EXECUTION]: The skill executes local tools and scripts to perform its intended development tasks.
Evidence: Invokes unity-cli for environment preflight and unity_test_batch.py for submitting automated test runs.
Context: Execution is governed by a lifecycle lock protocol and a mandatory preflight 'gate' to ensure repository readiness and ownership before mutation.

implement-task