The Agent Skills Directory

[COMMAND_EXECUTION]: The skill generates shell scripts (.sh) and uses the Bash tool to execute them. It includes instructions to make files executable via 'chmod +x' and encourages running them as the 'final acceptance gate'.
[COMMAND_EXECUTION]: User-provided $ARGUMENTS are used directly to determine file paths and directory names for script generation. If $ARGUMENTS contains an explicit path with '/' or special characters, the skill 'respects it as-is', which could lead to path traversal or unexpected file writes outside the intended '.more-loop' directory if platform-level sanitization is missing.
[REMOTE_CODE_EXECUTION]: The skill's instructions provide templates for 'Behavioral' testing that pipe the output of a curl command (fetching from '127.0.0.1') directly into a 'python3' interpreter for JSON processing. While targeting localhost, this 'piped to interpreter' pattern is inherently dangerous and flagged as a high-risk execution vector.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection. It reads local codebase files (via Read, Glob, Grep) to automatically generate test criteria, potentially ingesting malicious instructions embedded in the analyzed project that could influence the generated verification logic.
Ingestion points: Local codebase files read via Read, Glob, and Grep tools.
Boundary markers: Absent; the skill does not specify delimiters or 'ignore' instructions when processing codebase content.
Capability inventory: Write (to create scripts) and Bash (to execute them).
Sanitization: None; the skill lacks validation or escaping for codebase content before interpolating it into the logic of generated shell scripts.

more-loop-verify