The Agent Skills Directory

[COMMAND_EXECUTION]: The evaluation script evals/run_behavior_suite.py uses the subprocess module to run behavior tests through the codex CLI in a sandboxed environment.
[DYNAMIC_EXECUTION]: The evals/run_behavior_suite.py script utilizes importlib.util to load local project files for internal testing and grading logic.
[INDIRECT_PROMPT_INJECTION]: The skill processes workflow artifacts and traces, creating an ingestion surface for indirect injection. However, the skill mandates strict JSON schema validation and formal boundary markers as documented in the security and governance references. Mandatory Evidence Chain: 1. Ingestion points: GraphSpec JSON and run traces. 2. Boundary markers: Required schema_version and artifact-specific schemas. 3. Capability inventory: Static validation tools and sandboxed evaluation scripts. 4. Sanitization: Comprehensive schema-based validation for all artifact inputs.

graph-engineering