What happens if the content cannot be fixed automatically?

If the skill cannot reach a PASS status within three iterations, it stops and generates an actionable TODO report identifying exactly what is blocking the quality gate so the user can intervene.

Which file types can this skill polish?

It currently supports specific research deliverables including output/SNAPSHOT.md, TUTORIAL.md, SYNTHESIS.md, REVIEW.md, and IDEA_SHORTLIST.md.

What is a 'self-loop' in this context?

A self-loop is an automated iterative process where the skill evaluates its own output against a specific rubric, applies targeted fixes to failures, and re-evaluates the content until it meets the required quality bar.

How does the skill prevent hallucinations or 'hollow' writing?

The skill follows an 'evidence-first' guardrail, meaning it is prohibited from inventing results and must ground every claim in specific workspace inputs like extraction tables or core paper sets.

Does this skill require specific triggers?

Yes, it can be triggered by phrases such as 'self loop', 'polish deliverable', 'quality gate', or 'fix-on-fail' within the Claude Code environment.

Deliverable Self-Loop & Quality Gate

Name: Deliverable Self-Loop & Quality Gate
Author: WILLOSCAR

byWILLOSCAR

•

236

•

콘텐츠 관리

Refines and validates research deliverables through an iterative diagnose-fix-verify loop to ensure high-quality, evidence-backed output.

This skill acts as a deterministic quality gate for research pipelines, automating the process of polishing drafts into publication-ready documents. By implementing a structured 'diagnose -> fix -> re-check' workflow, it ensures that deliverables like systematic reviews, tutorials, and research snapshots meet rigorous standards. The skill cross-references outputs against internal data artifacts to prevent 'hollow' writing, ensuring every claim is grounded in evidence and every tutorial step is verifiable, making it an essential tool for maintaining high signal-to-noise ratios in technical and academic workflows.

주요 기능

01Evidence-first validation that enforces citations from core paper sets and extraction tables

02Scope-creep prevention by strictly aligning deliverables with locked upstream contracts and plans

03236 GitHub stars

04Automated reporting via DELIVERABLE_SELFLOOP_TODO.md with clear PASS/FAIL status and blockers

05Multi-rubric support for specific document types including Snapshots, Tutorials, and Syntheses

06Iterative self-correction logic that performs up to three loops to reach a PASS state

사용 사례

01Hardening a draft research snapshot by replacing generic narration with concrete, cited claims

02Ensuring a technical tutorial remains consistent with an approved module plan and decision log

03Validating a research idea shortlist against diversity, falsifiability, and minimal validation criteria

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add willoscar/research-units-pipeline-skills deliverable-selfloop

For use in Claude.ai and ChatGPT

Download Skill