The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill's reference documentation in 'references/esm-c-api.md', 'references/forge-api.md', and 'references/workflows.md' provides code examples that utilize the Python 'pickle' module for caching protein embeddings and checkpointing long-running batch jobs. The use of 'pickle' for deserialization is a known security risk as it can lead to arbitrary code execution if a user is tricked into loading a maliciously crafted pickle file.
[INDIRECT_PROMPT_INJECTION]: The skill possesses an attack surface for indirect prompt injection through the ingestion of untrusted biological data.
Ingestion points: Untrusted data enters the agent context through protein sequences and PDB structures processed in 'SKILL.md' and 'references/esm3-api.md'.
Boundary markers: No explicit delimiters or instructions to ignore embedded content are used in the provided code examples.
Capability inventory: The skill enables file system writes (PDB export), network operations via the Forge API, and complex model inference.
Sanitization: There is no evidence of sanitization for sequence data or metadata within structure files before they are processed or returned to the agent context.
[EXTERNAL_DOWNLOADS]: The skill requires the installation of external Python dependencies, specifically 'esm' and 'flash-attn', which are fetched from public package registries during the setup process described in 'SKILL.md'.

esm