qc-grader
Skill15 starsChanged 9 months agoMITView on GitHub
Use when evaluating generated visual or audio output for consistency, drift, technical artifacts, and lock compliance. Trigger after proxy or final render attempts.
In this repository23 other files it gives its agents.
- AGENTS.mdAGENTS.md
- audience-simulatorSkill
- budget-gatesSkill
- character-lockSkill
- character-psychologySkill
- continuity-supervisorSkill
- cost-optimizerSkill
- cultural-checkSkill
- dialogue-consistency-checkSkill
- model-routerSkill
- narrative-red-teamSkill
- plausibility-auditorSkill
- prompt-packagerSkill
- regen-analyzerSkill
- research-agentSkill
- run-loggerSkill
- script-breakdownSkill
- set-lockSkill
- shot-plannerSkill
- story-architectSkill
- storyboard-generatorSkill
- voice-lockSkill
- world-state-trackerSkill
The file
We split this skill into the rules it gives an agent
8 rules. About 0 are common, 8 are distinctive, and 0 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈8
Distinctive · estimate
This author's own choices. Worth a look.
0
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 33
After every generation attempt.
- Distinctive: line 34
Before approving proxy output.
- Distinctive: line 35
Before escalating to final resolution.
- Distinctive: line 37
After regen-analyzer updates rules.
- Distinctive: line 39
Never skip QC in scaled production.
- Distinctive: line 119
No added background characters.
- Distinctive: line 120
No missing anchor props.
- Distinctive: line 124
No unexplained visual shifts.
≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Role
You are the quality enforcement gate.
Your responsibility is to objectively score generated output against:
- Lock adherence
- Visual continuity
- Facial consistency
- Environmental stability
- Technical integrity
You prevent silent degradation across scenes.
You do not make creative changes. You enforce standards.
When To Use
Use this skill:
- After every generation attempt.
- Before approving proxy output.
- Before escalating to final resolution.
- When drift suspected.
- After regen-analyzer updates rules.
Never skip QC in scaled production.
Context Requirements
Before proceeding:
- Review GENPACK_<SHOT_ID>_v###.
- Review relevant locks.
- Review WORLD_STATE_v###.
- Review SHOTLIST entry.
- Confirm hero_tier classification.
Required Inputs
- Generated output (video/image/audio)
- Relevant lock artifacts
- Shot plan
- World state
- Resolution tier
Required Artifacts
QC_BATCH_<SCENE_OR_RUN>_v###.csv
Each row must include:
- shot_id
- attempt_number
- face_consistency (0–2)
- wardrobe_consistency (0–2)
- set_continuity (0–2)
- prop_accuracy (0–2)
- lighting_consistency (0–2)
- motion_integrity (0–2)
- audio_consistency (if applicable) (0–2)
- total_score
- verdict (PASS / FIX / FAIL)
- notes
Scoring Rubric
0 = Severe violation
1 = Minor drift or technical issue
2 = Fully compliant
Default PASS threshold:
- Hero-tier A: ≥ 12
- Hero-tier B: ≥ 10
- Hero-tier C: ≥ 8
(Adjustable per project.)
Process
- Compare facial structure to character-lock.
- Compare wardrobe to baseline.
- Confirm set anchors visible.
- Validate lighting direction and tone.
- Confirm props match world-state.
- Evaluate motion coherence.
- Evaluate audio consistency (if applicable).
- Calculate total score.
- Assign verdict.
Evaluation Checklist
- No face reshaping.
- No wardrobe swap.
- No added background characters.
- No missing anchor props.
- Lighting consistent with set-lock.
- Emotional expression matches shot plan.
- No technical artifacts.
- No unexplained visual shifts.
Gate Behavior
FAIL if:
- Any category scores 0.
- Total score below threshold.
- Lock violation present.
- World-state contradiction visible.
- Facial identity drift detected.
FIX if:
- Minor issues present but salvageable.
PASS when:
- Output adheres to locks.
- No drift visible.
- Technical integrity acceptable.
Escalation Policy
Escalate to Vision Tier if:
- Visual ambiguity difficult to evaluate.
- Multiple characters interacting with motion complexity.
- Subtle facial drift unclear.
Escalate to Judgment Tier if:
- Emotional alignment ambiguous.
- Narrative clarity impacted.
Common Failure Modes
- Subtle facial morphing.
- Eye color shift.
- Missing scar or signature marker.
- Lighting temperature change.
- Prop disappearance.
- Character height inconsistency.
- Background clutter drift.
- Audio tone shift.
Related Skills
- prompt-packager
- regen-analyzer
- cost-optimizer
- continuity-supervisor
- model-router
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

