The Agent Skills Directory

[COMMAND_EXECUTION]: The skill's core functionality relies on executing commands defined in an external JSON "bundle" (test-audit-bundle.json) to run the repository's test suite. These commands are executed via subprocess.Popen within an isolated temporary directory.
[COMMAND_EXECUTION]: The skill uses the git command-line utility to inspect the repository state and manage local exclusions. These calls are performed using subprocess.run with validated or sanitized inputs.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted JSON data from "bundle" and "audit" files, which defines the commands to be executed during the audit process.
Ingestion points: The skill ingests data from files specified by the bundle and audit input parameters in skill-protocol.json.
Boundary markers: The skill implements strict schema validation and SHA256 checksum verification for "sealed" audits in scripts/audit_contract.py.
Capability inventory: The skill can execute shell commands in an isolated copy of the repository and write output artifacts (test-audit.md) and git metadata (.git/info/exclude) to the target repository.
Sanitization: Path inputs are validated to prevent traversal outside the repository, and workspace slugs are normalized to alphanumeric characters.
[DYNAMIC_EXECUTION]: The skill uses importlib.import_module in scripts/_skill_protocol_runtime.py to load its internal diagnostic contract at runtime.
[SAFE]: The skill implements a "throwaway copy" strategy in scripts/replay.py to isolate code execution and mutations, ensuring that the original repository's source code remains unmodified during the audit process.

temper