The Agent Skills Directory

[SAFE]: The skill is restricted to mathematical reasoning and documentation. It explicitly prohibits the agent from generating code, which mitigates risks associated with executable payloads.
[INDIRECT_PROMPT_INJECTION]: The skill possesses a data ingestion surface by reading external reference files and orchestrator-provided problem specifications. However, the risk is minimal because the agent has no capability to execute commands or make network requests.
Ingestion points: Reference files in cfd-bioreactor/references/ (physics-models.md, validation-benchmarks.md) and orchestrator inputs (problem specification, proposed approach).
Boundary markers: Absent; the skill does not specify delimiters for external content.
Capability inventory: Reading local files via a 'Read' tool and writing YAML specifications via a 'Write' tool. No network, subprocess, or dynamic execution capabilities identified.
Sanitization: Absent; the skill does not explicitly describe sanitization of ingested data.
[DATA_EXPOSURE]: File access is limited to project-specific reference documentation and local session directories for handoff files. No access to sensitive system paths (e.g., .ssh, .aws, .env) or hardcoded credentials was found.

cfd-mathematician