The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest data from web searches and repository files. It mitigates the risk of processing malicious instructions by requiring the agent to treat all ingested data solely as untrusted evidence and by validating the output against a strict structural schema.
Ingestion points: Repository files and web search results are ingested during Source and Research phases.
Boundary markers: Explicit prompt instructions define ingested content as evidence rather than instructions.
Capability inventory: Includes file system writes within a scoped workspace and execution of local Git commands.
Sanitization: Workspace slugs are sanitized to prevent path traversal, and drafted content is validated for structural compliance.
[COMMAND_EXECUTION]: The skill uses subprocess.run to interact with Git and execute local validation scripts. These calls are securely implemented using shell=False and incorporate sanitized inputs, preventing command injection vulnerabilities.
[DYNAMIC_EXECUTION]: Dynamic imports (importlib.import_module) are used within the runtime to load diagnostic contracts, providing environment compatibility while keeping imports scoped to the package's own files.
[PERSISTENCE]: The skill manages research artifacts by updating the Git info/exclude file. This is a legitimate configuration task for local workspace management and does not constitute an unauthorized persistence mechanism.

ideate