The Agent Skills Directory

[SAFE]: The skill instructions strictly follow safe development practices, such as assertion preservation and minimal code changes, which avoids weakening software security during the test stabilization process.\n- [COMMAND_EXECUTION]: The skill uses npx vitest to execute tests locally. This is a standard and safe operation for verifying code changes in a development environment.\n- [PROMPT_INJECTION]: The skill processes untrusted issue data to target specific tests. Ingestion points: Issue body. Boundary markers: Absent. Capability inventory: Shell execution (npx vitest) and file system access (writing failure.md and e2e-report.md). Sanitization: Absent. The risk of indirect prompt injection is mitigated by the highly specific and limited set of allowed code modifications.

deflake