The Agent Skills Directory

[SAFE]: The skill instructions and associated reference materials are focused entirely on the stated goal of academic model replication. The coding standards provided (references/CODING_STANDARD.md) encourage best practices such as type hinting, tensor shape tracking, and explicit weight initialization.
[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface as it is designed to ingest and process external, user-provided academic papers (PDF/text). This surface is necessary for the skill's primary function.
Ingestion points: User-supplied research papers processed via the Read tool (SKILL.md).
Boundary markers: Absent; there are no specific instructions to the agent to treat segments of the paper as potentially untrusted data or to use delimiters.
Capability inventory: The skill is granted access to powerful tools including Write, Edit, and Bash (SKILL.md).
Sanitization: No explicit text sanitization or filtering is defined for the ingested research content.
[COMMAND_EXECUTION]: While the skill permits the use of the Bash tool, its use is intended for validating model shapes, checking parameter counts, and verifying gradient flow as part of the implementation workflow. No dangerous or unauthorized command patterns were found in the skill logic.

paper-replication