references/qc

Diagnose AI video generation failures and enforce delivery acceptance gates.

38|4|Updated Jul 29, 2026
One-click install
npx skills add https://github.com/6174/recut --skill references-qc-6174
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: references/qc
Source: https://github.com/6174/recut/tree/main/service/skills/recut-director/references/qc
Command: npx skills add https://github.com/6174/recut --skill references-qc-6174

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) and assets (resource) components.

What problem does it solve? AI-generated video clips often fail in recognizable ways—identity drift, melting geometry, missing actions, broken continuity—and creators lack a systematic way to name the cause, pick the cheapest fix, and decide whether a finished cut is actually deliverable. ## Core Features & Use Cases - Failure Diagnosis (F1–F19): Translates viewer complaints like "the face changed" or "it looks like a slideshow" into coded failure modes with ranked causes and cost-ordered repair ladders (L1 prompt edit through L7 cut the shot). - Three Acceptance Gates: Provides weighted pre-generation, per-clip post-generation, and assembled-sequence checklists with pass thresholds and blocker rules that veto delivery regardless of score. - Repair Loop Discipline: Enforces a three-strike rule—after three failed generations of the same shot design, change the design rather than rewording the prompt. - Use Case: A generated shot shows a sixth finger at 0:03.6. The skill codes it F10, identifies it as a blocker, and directs a first/last-frame regeneration or a shot split instead of blind retries. ## Quick Start Ask the agent to diagnose why the latest generated clip looks wrong and tell you the cheapest fix before regenerating.

Frequently Asked Questions about references/qc

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I fix AI video generation failures like face changes or melting?▼

Translate the symptom into a failure code first: face or wardrobe changes are F1 identity drift, melting geometry is F3 warping. Then apply the cost ladder from L1 prompt edits upward, changing only one variable per retry.

What is the cheapest way to repair a bad generated video clip?▼

Start at L1 with a prompt edit, then L2 parameter changes like shorter duration or lower motion strength. Only escalate to L4 keyframe rebuilds or L5 shot re-planning when lower rungs fail, and never skip levels without a documented reason.

When should I stop regenerating the same AI video shot?▼

After three failed generations of the same shot design, apply the three-strike rule: change the design by shortening, pushing in, simplifying, or splitting the shot. A fourth prompt rewording rarely produces new information.

How do I check if an AI-generated video is ready to deliver?▼

Run three gates: a pre-generation plan check, a per-clip post-generation score, and a sequence gate on the assembled cut requiring 85/100. Any blocker such as identity breaks, extra fingers, or vanished props vetoes delivery regardless of score.

Why does my AI video look like a slideshow with no motion?▼

That symptom is F2, caused by prompts describing a picture rather than an event, static keyframes, or motion strength set too low. Rewrite the prompt with body, cloth, and environment motion layers and rebuild the keyframe mid-gesture.

What AI video failures cannot be fixed by regenerating?▼

Irreducible failures include sustained precise hand-object contact, readable long text, multi-person contact, exact mirroring, and recognizable crowds. These must be rewritten at the storyboard stage as intent-to-offscreen-to-result or replaced with stills, practical footage, or compositing.