The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses subprocess.run extensively to coordinate the installation and execution of Node.js binaries and Python scripts across different domain plugins (e.g., in scripts/bootstrap.py, scripts/workflow.py, and scripts/setup.py). These executions are strictly scoped to the managed runtime environment and are used to invoke verified toolchains.
[EXTERNAL_DOWNLOADS]: The skill fetches the Node.js runtime from nodejs.org and domain-specific plugin bundles from the author's GitHub repositories (full-aigc-skills and full-aigc-plugins).
Integrity is maintained via mandatory SHA256 verification against hardcoded lockfiles (node.lock.json and distribution.lock.json).
Download logic in scripts/setup.py includes a strict URL regex whitelist that prevents fetching content from unauthorized domains or repositories.
[DYNAMIC_EXECUTION]: scripts/native_contract.py implements a sophisticated security wrapper that intercepts Python module loading at runtime.
It injects a StrictSession validator that checks tool input schemas and command catalogs against known-safe snapshots before permitting execution.
This mechanism effectively prevents the execution of undocumented commands or those that have drifted from their verified security baseline.
[INDIRECT_PROMPT_INJECTION]: The skill processes external JSON-formatted workflow plans that define task sequences.
Ingestion points: scripts/workflow.py and scripts/brief.py ingest plan and brief files from the filesystem to drive orchestration.
Boundary markers: scripts/brief.py implements rigorous schema validation, type enforcement, and dependency cycle detection to ensure inputs conform to expected structures.
Capability inventory: The skill possesses the ability to execute shell commands and write to local project directories as part of its primary orchestration function.
Sanitization: Input data is validated against a strict JSON schema and integrity-checked via hashes before influencing any system-level operations.

artcraft-harness