The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The file scan_demo.sh contains hardcoded session cookies in the COOKIE variable, including sensitive tokens such as pinId, TrackID, and 3AB9D23F7A4B3CSS. While these are used in a demonstration context targeting the vendor's own domain (jd.com), hardcoding authentication credentials or session tokens in scripts is a security risk that can lead to session hijacking.
[COMMAND_EXECUTION]: The script verify.sh programmatically constructs and executes curl commands based on user-supplied parameters (-u, -c, -m, etc.). Although the script uses safe shell array expansion ("${CMD[@]}") to prevent shell injection, it provides a functional capability for the agent to execute arbitrary network requests.
[EXTERNAL_DOWNLOADS]: The skill uses curl to fetch content from remote URLs for vulnerability analysis. The implementation in verify.sh includes the -k (insecure) flag, which explicitly disables SSL/TLS certificate verification, potentially exposing the operation to man-in-the-middle attacks.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from remote HTTP responses (headers and body) and processes it using pattern matching to identify vulnerabilities. This represents a surface for indirect prompt injection if the processed data were to influence downstream agent instructions.
Ingestion points: The verify.sh script captures the output of remote network requests into the RESPONSE variable.
Boundary markers: None identified; the script performs pattern matching directly on the raw response string.
Capability inventory: The skill uses curl for network access and grep for parsing response content.
Sanitization: The script does not validate or sanitize the remote content before analyzing it for success criteria patterns.

pentest-vuln-verify-test