The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill setup involves downloading the mfapy package from an external Git repository during installation.
Evidence: Both SKILL.md and assets/requirements.txt point to https://github.com/fumiomatsuda/mfapy.git at a specific commit hash.
[DYNAMIC_EXECUTION]: The underlying mfapy engine dynamically generates and executes numerical functions to solve metabolic flux equations based on the provided network model.
Evidence: scripts/_mfa_model.py documents this internal behavior and includes verification logic (hasattr(self._engine, "func")) for the generated numerical functions.
[INDIRECT_PROMPT_INJECTION]: The skill ingests scientific model and measurement data from JSON files, creating a potential attack surface for indirect prompt injection if malicious instructions were embedded in metadata or strings.
Ingestion points: model.json and measurements.json are loaded and processed in scripts/mfa.py.
Boundary markers: The skill does not use explicit boundary markers or "ignore instructions" warnings in the processing logic for external data.
Capability inventory: The skill scripts are restricted to local data processing, optimization via SciPy, and file writing. No shell execution or network operations are performed by the scripts themselves.
Sanitization: The skill mitigates risks by validating all identifiers (metabolites, reactions, fragments) against a strict alphanumeric regex ([A-Za-z][A-Za-z0-9]{0,39}) before they are passed to the numerical modeling engine.

13c-metabolic-flux