The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local validation and testing scripts.
Evidence: Step 5 in SKILL.md executes node <this-skill>/scripts/check-skill.mjs and Step 7 executes <this-skill>/scripts/fires.sh.
Evidence: scripts/fires.sh utilizes the claude CLI tool to simulate and verify trigger phrases for new skills.
[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied requests and procedure descriptions, which serves as a surface for potential instruction injection in the generated output.
Ingestion points: User interview responses (Step 1) and the content of the target skill folder (Step 5 and Step 7).
Boundary markers: The skill instructs the agent to use structured templates and mechanical scripts, though it lacks explicit isolation markers for every interpolation point.
Capability inventory: The skill possesses file system access and local script execution capabilities.
Sanitization: scripts/check-skill.mjs validates the structural integrity and frontmatter but does not perform semantic safety filtering of the generated content.
[DYNAMIC_EXECUTION]: The skill facilitates the generation and execution of scripts based on provided templates.
Evidence: references/skill-template.md provides templates for Bash scripts that the agent is prompted to create and execute.
Evidence: Internal Node.js and Bash scripts are invoked at runtime to manage the development and verification workflow.
[SAFE]: The skill includes robust security and safety instructions.
Evidence: SKILL.md and references/review-checklist.md explicitly mandate the avoidance of hardcoded secrets and require human approval for system-impacting actions.
Evidence: scripts/fires.sh proactively disables dangerous tools (Write, Edit, Bash) using the --disallowedTools flag when performing trigger testing simulations.

write-great-skill