The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes common command-line tools including curl, grep, jq, and python3 to automate reconnaissance tasks. This behavior is consistent with the tool's goal of identifying attack surfaces on SMB websites.
[EXTERNAL_DOWNLOADS]: Retrieves certificate information from crt.sh. This is a well-known service for certificate transparency logs and is used here for legitimate domain discovery purposes.
[PROMPT_INJECTION]: The skill identifies an indirect prompt injection surface as it processes untrusted data from target websites (e.g., WordPress debug logs) and external APIs.
Ingestion points: Remote data is fetched from target websites and crt.sh using curl commands in the SKILL.md file.
Boundary markers: No delimiters or instructions are provided to the agent to ignore potentially malicious content within the processed files.
Capability inventory: The skill utilizes subprocess execution for several CLI tools (curl, python3, grep, jq) across its instructions.
Sanitization: No sanitization or validation of the external content is performed before it is handled by the agent or piped to local utilities.
[DATA_EXFILTRATION]: While the skill extracts potentially sensitive data like customer PII from debug logs on target sites, it stores this information in local files (e.g., /tmp/carpet_debug.log) for the user's manual review and does not transmit it to external attacker-controlled servers. The use of evil.com in headers is a standard technique for testing CORS reflection vulnerabilities.

recon-carpet-cleaning