The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes npx shiplight test to run the test suite and verify fixes. This is the intended primary function of the tool for test reproduction and verification.
[PROMPT_INJECTION]: The skill possesses an attack surface for indirect prompt injection as it ingests untrusted data from the application under test, including DOM structures via inspect_page, browser console logs, and network logs. This content could theoretically contain instructions designed to influence the agent's diagnosis or the YAML fixes it generates.
Ingestion points: Browser session data (DOM, logs, network) accessed in Phase 3.
Boundary markers: None explicitly defined in the instructions for ingested web content.
Capability inventory: Shell execution (npx shiplight), browser interaction (act), and file writes (modifying .test.yaml files).
Sanitization: Not specified; the agent relies on its internal logic to interpret the UI state.
[SAFE]: The skill's file-writing operations are restricted to updating existing YAML test files and generating markdown/HTML reports in a local directory (shiplight-report/), which is standard behavior for testing tools.

triage