The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process data from the application under test (API responses, frontend source code, and data layer outputs). It uses this data to populate a qa-report.json and qa-report.md. Since it possesses the capability to execute shell commands via Bash and curl to interact with services, a malicious payload in the application data could potentially influence the agent's behavior during the reporting phase if boundary markers are not strictly observed.
Ingestion points: contracts/ directory, service endpoints via curl, and source code analysis using grep (as seen in references/validation-checklist.md).
Boundary markers: The instructions mandate schema conformance and use of specific rubrics, but do not explicitly define delimiters for untrusted data in the prompt interpolation process.
Capability inventory: Uses Bash, curl, and python3 to execute tests and verify service status. Also uses Glob, Grep, Read, and Write for file system operations.
Sanitization: The skill instructions emphasize machine-readable schema validation for the output (qa-report-schema.json), but do not detail sanitization of the input data fetched during Phase 3 adversarial probing.

qe-agent