The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes project test suites and maintenance commands (e.g., clearing caches or reinstalling dependencies) via the Bash tool. These actions are governed by a safety gate that requires user confirmation for any operation with side effects and blocks destructive actions entirely.
[DATA_EXFILTRATION]: No exfiltration patterns were detected. The skill implements a mandatory 'Secret Redaction' pipeline (documented in references/triage-pipeline.md) that uses high-entropy regex patterns to scrub API keys, private keys, and tokens from test logs before they are processed by the LLM or stored in cache files.
[PROMPT_INJECTION]: The triage pipeline processes external test output which represents an indirect prompt injection surface. This is mitigated by several controls: 1. Ingestion points: Test runner stdout/stderr is parsed in references/triage-pipeline.md; 2. Boundary markers: The LLM prompt uses clear markdown headers and JSON formatting to separate data from instructions; 3. Capability inventory: The skill utilizes Bash, Write, and Agent tools; 4. Sanitization: A mandatory secret redaction step is applied to all ingested logs to prevent credential leakage or control character exploitation.

test-deep