The Agent Skills Directory

[COMMAND_EXECUTION]: The documentation suggests bypassing SSL certificate verification by setting the environment variable NODE_TLS_REJECT_UNAUTHORIZED=0. This practice disables standard security protections and increases susceptibility to Man-in-the-Middle (MitM) attacks.
[INDIRECT_PROMPT_INJECTION]: The curl_to_bru Python script in gpt.md constitutes a vulnerability surface because it processes untrusted input to perform file system operations.
Ingestion points: The script accepts a curl_command string and an output_path from the execution context.
Boundary markers: There are no instruction-level boundary markers or delimiters defined to prevent the agent from misinterpreting embedded commands within the input string.
Capability inventory: The script performs directory creation via os.makedirs and writes files using open(..., 'w') in gpt.md.
Sanitization: No input validation or path sanitization is performed, which could allow writing to sensitive locations if the agent is directed by untrusted input.

bruno