The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes a local Python script located in the skill's directory to manage test walkthroughs and reports. It also performs 'server-side probes' which include database queries, file system access, and API calls as specified in the test guide's backend instructions.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection as it follows natural language instructions provided in an external JSON guide. This file dictates browser actions and backend probes, which could be exploited if the guide is poisoned. * Ingestion points: .skills//review-product-flow.json (Try, Expect, and backend fields). * Boundary markers: Uses structured JSON fields for step definitions but lacks internal sanitization of the commands or instructions within those fields. * Capability inventory: Browser automation via Playwright/Chromium, shell execution of local scripts, and file/database/API access for verification. * Sanitization: The skill enforces a mandatory check for a fresh 'vet-product-flow' report, which acts as a security gate to validate the guide before execution.
[DATA_EXFILTRATION]: The skill includes a server command that binds to a local port (127.0.0.1:8787). While limited to the local interface, this establishes a network service capable of serving data from the run file during execution.

run-product-walkthrough