The Agent Skills Directory

[SAFE]: The skill acts as a project manager and does not perform substantive data processing itself; it dispatches work to specialized sub-skills (deep-research, academic-paper, etc.), maintaining a clear separation of concerns.
[SAFE]: Implements a mandatory 'Integrity Review Protocol' (Stages 2.5 and 4.5) that uses WebSearch and DOI lookups to verify 100% of references and data claims against real-world sources before proceeding, which is an exemplary security and accuracy feature.
[PROMPT_INJECTION]: The orchestrator uses 'MANDATORY' and 'CRITICAL' instructional keywords to enforce its state machine and safety gates. These are legitimate control flow mechanisms and do not attempt to bypass platform safety guidelines.
[COMMAND_EXECUTION]: The skill utilizes 'tectonic' and 'pandoc' in Stage 5 and Stage 6 to compile LaTeX documents into PDFs. This is standard functionality for an academic paper orchestrator and is executed in a controlled manner within the documented workflow.
[SAFE]: Includes an 'Adaptive Checkpoint System' that forces user confirmation at critical junctions (integrity failures, review decisions, and finalization), preventing the agent from performing significant actions without explicit human oversight.
[SAFE]: Employs a 'State Tracker Agent' as a single source of truth for version control and material history, providing a transparent audit trail of all Human-AI collaboration steps.

academic-pipeline