The Agent Skills Directory

[SAFE]: The skill implements a sophisticated academic workflow controlled by strict 'IRON RULES' that enforce mandatory integrity checks and human-in-the-loop checkpoints, preventing autonomous execution of potentially dangerous operations.
[SAFE]: Data exfiltration risks are mitigated via mandatory consent gates in the integrity and collaboration depth agents, requiring explicit user approval before sharing dialogue logs or metadata with external models for cross-verification.
[SAFE]: The system utilizes a 'Material Passport' for session persistence, which uses cryptographic hashing and local file locking (fcntl.flock) to maintain a secure and transparent research audit trail.
[SAFE]: Indirect prompt injection is addressed by using clear data delimiters in agent prompts (e.g., 'CLAIM: {claim_text}') to ensure that user-provided research text is processed as data rather than instructions.
[SAFE]: External dependency management is transparent, relying on well-known academic tools like Pandoc and Tectonic for document formatting and reputable services such as Semantic Scholar for reference verification.

academic-pipeline