The Agent Skills Directory

[COMMAND_EXECUTION]: Automates complex reconnaissance tasks using shell loops and variables ($TARGET). It uses piped commands to process output from remote services, including the execution of local Python scripts to parse JSON and XML data.
[DATA_EXFILTRATION]: Specifically targets sensitive file paths and directories identified as containing PII and financial data. Evidence includes instructions to search for directories such as /wp-content/uploads/credit-apps/ (explicitly noted to contain SSNs and income data) and debug.log (containing customer contact information, vehicle VINs, and SQL queries).
[EXTERNAL_DOWNLOADS]: Fetches domain discovery data from crt.sh, a well-known service for certificate transparency logs. This data is piped into a local Python script for parsing and deduplication. This is a standard reconnaissance pattern and uses a well-known service.
[PROMPT_INJECTION]: The skill exhibits a surface for indirect prompt injection by ingesting untrusted data from the target's web server into the agent's context without sanitization.
Ingestion points: Reads content from debug.log, WordPress user API (/wp-json/wp/v2/users), and XML-RPC responses.
Boundary markers: None identified; untrusted content from the target server is processed and displayed alongside agent instructions.
Capability inventory: Includes the ability to execute shell commands (curl, jq, tee) and Python scripts (python3 -c).
Sanitization: No evidence of input validation or sanitization before processing remote data.

recon-automotive-dealers