The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: Automated scans flagged a command in SKILL.md involving curl piped to python3. Investigation confirms this is a false positive; the full command curl -s https://api.example.com/data | python3 -m json.tool invokes a built-in Python module to pretty-print JSON responses and does not execute the fetched data as a script.
[INDIRECT_PROMPT_INJECTION]: The provided helper scripts (scripts/api-tester.sh and scripts/http-debug.sh) accept user-provided strings for URLs, methods, and data, which are then used in shell commands.
Ingestion points: Script arguments such as $url, $method, and $data passed to bash functions.
Boundary markers: Absent; the skill does not include instructions or logic to validate the safety of these inputs.
Capability inventory: Subprocess execution of curl, jq, and dig for network and data operations.
Sanitization: The scripts utilize double quotes around variables, which offers basic protection against simple whitespace-based shell injection but does not prevent more advanced injection techniques if the agent provides malicious input.

curl-http