The Agent Skills Directory

[SAFE]: The skill implements mandatory 'Integrity Gates' at Stages 2.5 and 4.5, which use specialized agents to perform 100% verification of references, citations, and data accuracy. This process is designed to detect and block fabricated content and hallucinations, adhering to a 'zero tolerance' policy for integrity issues.
[SAFE]: An 'Adaptive Checkpoint System' is used throughout the pipeline. This system enforces 'IRON RULES' that prevent the AI from proceeding past critical stages (such as integrity checks or review decisions) without explicit user confirmation, ensuring human oversight at all mandatory boundaries.
[SAFE]: Subprocess execution is restricted to well-known, established academic tools such as Pandoc (for document conversion) and Tectonic (for LaTeX-to-PDF compilation). These tools are used for their intended purposes within the drafting and finalization stages.
[SAFE]: The 'Claim Reference Alignment Audit Agent' (v3.8) provides a specialized audit of claim faithfulness. It evaluates prose claims against retrieved source excerpts using a deterministic pipeline and structured judge prompts, mitigating the risk of indirect prompt injection from untrusted academic sources by enforcing evidence-bound verdicts.
[SAFE]: The skill uses a 'Material Passport' system for state management and reproducibility. This system tracks the provenance and verification status of all artifacts across sessions, ensuring that state transitions are based on verified data and that resumed sessions maintain integrity requirements.

academic-pipeline