The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill downloads the commitshow package from the NPM registry at runtime using npx.
[REMOTE_CODE_EXECUTION]: The npx commitshow@0.3.23 command executes external code locally. This process has access to the user's filesystem and environment variables. The skill attempts to mitigate risk by pinning the execution to a specific version.
[COMMAND_EXECUTION]: The skill executes shell commands (mkdir, npx, jq) using the repository path or a user-provided URL. This presents a potential command injection vector if the URL input is not properly sanitized before being passed to the shell.
[INDIRECT_PROMPT_INJECTION]: The skill parses and displays findings from an external audit engine, which could be manipulated by malicious data in the repository or from the remote API.
Ingestion points: Findings are read from .commitshow/audit.json generated by the external CLI.
Boundary markers: The instructions advise the agent to treat engine concerns as "untrusted advisory data" and to corroborate findings with local files.
Capability inventory: The skill allows the agent to read files, generate patches, and execute commands (with user approval).
Sanitization: No explicit sanitization or schema validation is performed on the descriptive text bullets before they are echoed to the user.

production-audit