The Agent Skills Directory

[COMMAND_EXECUTION]: The skill requires the agent to execute the newly created skill for verification purposes as described in steps/verify.md. This involves running instructions or scripts generated during the session.
[DYNAMIC_EXECUTION]: The skill facilitates the generation of instructions and executable scripts (within the scripts/ directory) which are then saved to the local file system and executed. This 'generate-and-run' pattern is an inherent dynamic execution vector.
[INDIRECT_PROMPT_INJECTION]: The skill's design phase involves reading 'raw material' and documentation from the repository to inform its instructions, which could lead to the ingestion of malicious content if the repository is poisoned.
Ingestion points: The skill reads repository files and user descriptions in steps/understand.md to establish the skill's purpose and material.
Boundary markers: The skill employs a 'recommend-then-confirm' format throughout steps/interview.md and steps/design.md, ensuring the user reviews and confirms every generated recommendation.
Capability inventory: The skill has the capability to write files and folders to local storage (steps/save.md) and execute skills for verification (steps/verify.md).
Sanitization: There is no explicit sanitization process mentioned for repository content before it is processed into new skill instructions.

bm-skill-builder