The Agent Skills Directory

[PROMPT_INJECTION]: The skill evaluates external file content from the local filesystem to determine feasibility, creating an indirect prompt injection surface.
Ingestion points: Data enters the context via Read, Glob, and Grep tools targeting files in the project workspace.
Boundary markers: The SKILL.md includes 'Iron Laws' that define the agent's logic and output format, but there is a lack of explicit delimiters or instructions to ignore commands embedded within the ingested data.
Capability inventory: The skill possesses the ability to read files and call other skills (via the Skill tool), though it is explicitly restricted from performing artifact creation itself.
Sanitization: There is no evidence of content sanitization or instruction filtering for the data read from the filesystem.
[COMMAND_EXECUTION]: The skill package includes local Node.js scripts (scripts/main.cjs, hooks/pre-execute.cjs, and hooks/post-execute.cjs). These scripts are used for basic lifecycle management and argument parsing. They utilize standard built-in modules like fs and path and do not contain patterns for unsafe command execution, subprocess spawning from user input, or dynamic code evaluation.

creation-feasibility-gate