The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The automated scanner flagged a command for piping remote content to Python. However, the command in SKILL.md (curl -s https://api.example.com/users | python -m json.tool) uses the standard library's json.tool module to pretty-print JSON data. It does not execute the downloaded content as a script, making this a false positive.
[COMMAND_EXECUTION]: The skill uses the terminal tool to execute curl, nslookup, and openssl. These are standard utilities for diagnosing API connectivity, DNS resolution, and TLS certificate validity.
[DYNAMIC_EXECUTION]: The skill employs execute_code for multi-step Python logic, including JWT payload decoding and implementing exponential backoff for rate limiting. These operations are performed within a restricted execution environment.
[INDIRECT_PROMPT_INJECTION]: The skill ingests data from external API endpoints via requests. While this constitutes an attack surface for indirect prompt injection, the skill includes explicit instructions for schema validation (validate_user function) and response content-type checking to mitigate risks from untrusted data.
[CREDENTIALS_UNSAFE]: The instructions demonstrate handling authentication tokens but correctly advise against hardcoding them. It recommends reading secrets from environment variables and provides a redact_auth utility function to prevent sensitive headers from being logged.

rest-graphql-debug