The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local bash scripts (e.g., project-state.sh, set-stage.sh, config-set.sh) to read and write project configuration. These operations are restricted to the vendor-provided scripts in the .claude/scripts/ directory as defined in the allowed-tools frontmatter.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted project data, such as design documents and existing code files, to infer the appropriate project stage and development rigor.
Ingestion points: Files such as project.yaml, design/gdd/game-concept.md, and design/game-brief.md are read into the agent's context.
Boundary markers: The skill does not implement explicit delimiters or warnings to ignore embedded instructions in the processed design documents.
Capability inventory: The skill can execute specific local scripts for configuration and invoke other internal project skills (e.g., /prototype, /brainstorm).
Sanitization: The agent maps inferred values to a strict set of predefined options for project stages (e.g., Concept, Systems Design) and rigor levels (e.g., minimal, standard), limiting the potential for arbitrary configuration injection.

start