The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied mission briefs and external specifications which drive the generation of implementation tickets and code. This creates a surface where instructions embedded in project documentation could influence agent behavior.
Ingestion points: Mission briefs, project README.md, CLAUDE.md, and supplied specification files.
Boundary markers: Uses a 'yard gate' audit to verify the environment and requires human approval of objectives and criteria (C1, C2) before implementation.
Capability inventory: The skill writes to multiple project files and executes code implementation via delegated workers.
Sanitization: Relies on a multi-stage human-in-the-loop process where users must approve the mission, criteria, and final output.
[COMMAND_EXECUTION]: The skill executes the drydock --check command to audit the repository state and spawns 'Team' workers to perform implementation tasks, builds, and test verification.
[DYNAMIC_EXECUTION]: During the closeout phase, the skill can generate and save new scripts or update the .mcp.json configuration based on 'lessons learned' during the run. This capability is strictly gated by human approval (C5) and a final verification pass.

launch