The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes local command-line utilities to perform its analysis and repair functions. It executes git for version control diffing and state management, and invokes project-specific test runners such as vitest, jest, pytest, and go test. It also calls build and linting tools like tsc, go build, mypy, and cargo check to verify the integrity of refactored code.- [DYNAMIC_EXECUTION]: During its mutation check phase, the skill generates temporary stub functions (e.g., returning null or zero) and writes them to production files to simulate logic failure. This represents a dynamic modification and execution of code strings, which is used here for the purpose of verifying that tests are sensitive to logic changes.- [INDIRECT_PROMPT_INJECTION]: The skill ingests and processes source code and test file content from the local project, which serves as a potential surface for indirect prompt injection. The skill addresses this risk through deterministic static analysis patterns and a mandatory confidence evaluation requirement (gated at 90%) before any autonomous code modification is performed.- [SAFE]: The skill implements industry-standard safety measures for its modification tasks, including the use of mktemp and cp for non-destructive file backups instead of risky stashing operations. It also enforces a strict restoration routine that compares the post-run state to a recorded pre-diff to ensure the working tree is returned to its original state regardless of the outcome of the test run.

test-provenance-guard