The Agent Skills Directory

[COMMAND_EXECUTION]: The skill directs the agent to execute multiple CLI commands using npx, including codeceptjs, codeceptq, and playwright. These are vendor-owned or established tools within the testing ecosystem used for test execution and artifact querying.
[REMOTE_CODE_EXECUTION]: The skill utilizes a run_code tool that allows for the execution of arbitrary CodeceptJS/JavaScript strings within a live browser session. This dynamic code execution is a core feature for interactive debugging but constitutes a potential risk if manipulated.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection due to its processing of external test data.
Ingestion points: The agent is instructed to read and analyze Markdown traces (trace.md) and HTML snapshots (*_page.html) generated during test execution in the output/ directory.
Boundary markers: The instructions do not define clear delimiters or specify how to distinguish between administrative instructions and content from external test artifacts.
Capability inventory: The agent possesses capabilities to execute code (run_code), run tests (run_test), and capture state (snapshot) within the testing environment.
Sanitization: There is no mention of sanitizing or validating the content of ingested HTML or trace files before analysis or subsequent tool use.

debugging-codeceptjs-tests