WILLOSCAR/research-units-pipeline-skills

What it solves

Research Harness provides infrastructure to make agent-assisted research inspectable, resumable, and honest. It prevents the common issue where an AI agent produces a final result (like a PDF) but leaves no trace of the evidence, decisions, or intermediate steps used to reach that conclusion, making it difficult to verify or resume work.

How it works

The system turns a research goal into a "Run" organized by a pinned Pipeline contract. It breaks tasks into recoverable Units and Skills, ensuring that every transition is backed by evidence and artifacts.

Key mechanisms include:

• Pinned Contracts: Snapshots the pipeline and skill implementations to prevent silent drift during a run.
• Evidence-Backed Completion: Success is not just a status flag; it requires matching artifact hashes, manifests, and completion events.
• Bounded Diagnosis: When a contract fails, the system identifies the specific repair surface rather than rewriting the entire process.
• Human Checkpoints: Allows users to approve specific artifacts (like outlines or protocols), which revokes authorization if the underlying artifact is changed.

Who it’s for

Researchers and developers using AI agents to conduct literature reviews, synthesize studies, or write technical reports who need a verifiable audit trail and the ability to recover from failures.

Highlights

• Diverse Workflow Templates: Includes pre-built workflows for research briefs, paper reviews, evidence synthesis, arXiv surveys, and tutorial creation.
• Harness Lock: Ensures execution integrity by hashing the kernel and skill bundle.
• Strict Residue Checking: Includes specific gates to prevent AI-generated template residue (e.g., "this run") from appearing in final deliverables.
• File-First Approach: Organizes all goals, decisions, and evidence into a structured workspace for full transparency.