The Agent Skills Directory

[COMMAND_EXECUTION]: The skill incorporates a suite of shell scripts (fork.sh, build-mutants.sh, select-mutant.sh, verify-mutant.sh, clean.sh, sync-patches.sh) to automate the mutation testing lifecycle. These scripts are designed with robust environment isolation in mind, using a throwaway 'fork' directory for all modifications. They utilize standard system tools such as rsync, git, and docker-compose to build and verify mutated container images without impacting the host repository or global system state.
[EXTERNAL_DOWNLOADS]: The skill references snouty, a CLI tool provided by the vendor (antithesishq), and accesses official documentation from vendor-controlled domains. These resources are standard components of the Antithesis testing ecosystem and are documented neutrally as prerequisites for the skill's operation.
[INDIRECT_PROMPT_INJECTION]: The skill exhibits an attack surface for indirect prompt injection as it ingests external data to perform its functions.
Ingestion points: Data enters the skill's context through snouty runs --json properties results, counterexample logs retrieved via snouty runs --json logs, and local source code scans during catalog reconstruction.
Boundary markers: The skill relies on structured JSON output from its own tools and specific unique log prefixes (e.g., ANTITHESIS MUTANT ACTIVE: {id}) to delineate test markers.
Capability inventory: The skill's scripts possess the ability to perform file writes (to the scratchbook, fork, and patch directories), execute git commands, and trigger Docker builds.
Sanitization: The skill employs structured parsing using jq for JSON data and targeted grep patterns to extract and validate specific markers, minimizing the risk of interpreting untrusted data as instructions.

antithesis-mutation-testing