The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill dynamically executes scripts found within the project repository.
It attempts to run node scripts/review-state.js and .claude/scripts/review-state.js to check the current gate state.
It checks for the existence of .claude/scripts/verify-runner.js using Glob and executes it with node if found.
[INDIRECT_PROMPT_INJECTION]: The skill processes and interprets data from external tool outputs, creating an attack surface.
Ingestion points: Reads output from scripts/review-state.js and various linter/test runners (e.g., ruff, pytest, cargo, go test).
Boundary markers: None are specified to distinguish between tool output and instructions.
Capability inventory: The skill has access to a wide range of shell tools (node, npm, python, cargo, go, gradle, etc.) and file system tools (Read, Grep, Glob).
Sanitization: There is no evidence of sanitization or escaping of the output received from these tools before it is processed by the agent.
[COMMAND_EXECUTION]: The skill executes a broad set of ecosystem-specific commands based on the detected project type (Node.js, Python, Rust, Go, Java). It interpolates $ARGUMENTS directly into the node .claude/scripts/verify-runner.js command, which could lead to command injection if the user-supplied arguments are not properly handled by the agent platform.

verify