The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from target server responses and stores it in a report without sanitization or proper boundary markers.
Ingestion points: In scripts/agent.py, the test_race_condition function captures the first 200 characters of the target server's response body (resp.text[:200]).
Boundary markers: The captured content is stored directly in a list and then serialized to a JSON report using json.dump without any delimiters to distinguish untrusted content from the report structure.
Capability inventory: The skill has the capability to perform network requests (requests) and write files to the local system.
Sanitization: No validation, filtering, or escaping is applied to the captured response text before it is recorded in the final output.
[DATA_EXFILTRATION]: The skill transmits user-provided authentication tokens (Bearer tokens) and application payloads to arbitrary external domains during testing. Additionally, the scripts/agent.py script explicitly disables SSL certificate verification by using verify=False and suppressing InsecureRequestWarning, which exposes the transmitted sensitive data to potential Man-in-the-Middle (MitM) attacks.
[COMMAND_EXECUTION]: The SKILL.md file contains several curl command templates for manual execution by the user to test business logic flaws such as price manipulation and workflow bypasses. These commands require the user to provide their own authorization tokens and target URLs.
[SAFE]: The skill refers to established and reputable security documentation from organizations like OWASP and PortSwigger, and the external URL references are for standard documentation purposes.

testing-for-business-logic-vulnerabilities