The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses git commands (git diff, git log) to identify modified files for determining testing scope. These use static, non-malicious arguments for local repository analysis.
[DATA_EXFILTRATION]: The instructions include a mandatory password redaction rule using [REDACTED], demonstrating a security-first approach to documenting authentication flows.
[PROMPT_INJECTION]: Ingesting untrusted content from web applications creates a surface for indirect prompt injection. The skill's instructions strictly limiting behavior to "reporting only" serve as a significant mitigation.
Ingestion points: External HTML, JavaScript console logs, and network traffic analyzed during testing.
Boundary markers: Not present for untrusted web content.
Capability inventory: Bash, Read, Write (for local reports), and WebSearch.
Sanitization: Not present.

qa-only