Skip to content
cccskills
FreePublished on GitHub

viz-frame-audit

Trigger when: (1) the user mentions "noise", "slop", "frame audit", "video review", "letter spacing", "padding intrusion", "frame extract", "naive arrow", or Korean equivalents (노이즈, slop, 프레임 검수, 글자 깨짐), or (2) after rendering a new or modified Manim scene, or (3) the user asks to review an mp4 in `media/videos/` or `~/Downloads/`. Post-render audit workflow for Manim 1080p60 videos. Extracts keyframes via ffmpeg, inspects them through Claude Code's Read tool, and classifies any defects into four standing categories (naive arrow / padding intrusion / glyph kerning drift / frame-order error). Catalogues 12+ verified incidents across the four GEODE scenes so each is caught the first time, not rediscovered. The companion authoring skill is [[manim-scene-craft]].

The English summary is being prepared. Showing the original description.

mangowhoiscloud/geode
See how to install

Files included (7)

  • SKILL.md5.0 KB
  • references/defect-catalogue.md5.1 KB
  • references/external-oss.md3.4 KB
  • rules/audit-workflow.md4.5 KB
  • rules/pixel-ratchet.md2.9 KB
  • rules/reporting.md2.9 KB
  • rules/typography-drift-gate.md5.0 KB

SKILL.md (original)

Review the instructions the agent will receive before you install.

Viz Frame Audit

Use after a render or re-render to verify the four defect categories. Skip it when only the deterministic layout --static-check is needed.

The four categories

Videos are inspected against the four standing defect categories below. The catalogue of verified incidents per category lives in references/defect-catalogue.md.

#CategoryDefinitionDetection signal
1Naive arrowHead too small, head colour mismatch with body, label crossing the dashed linehead_size ≤ 0.24, head_color ≠ body color with near-vertical arrow + label at UP * 0.25
2Padding intrusionText or box within 0.1 of another box edge, or crossing the canvas edgeBox height < content height, row labels at LEFT * 6.0+, 7+ outline lines in a small box
3Glyph kerning driftHelvetica Neue + Pango misreads specific glyph pairs ("GE ODE", "g eneration", "fit ness", "cr itic")Spurious space inside a word, more pronounced at smaller font_size
4Frame-order errorEmpty / half-empty transition frame, content not accumulating, line endpoint missing the last dot_clear_section + _set_section_title called separately, LaggedStart then separate Create(connectors)

Workflow

  1. Decide timestamps — cover the changed sections and adjacent transitions; use existing key-frame manifests when applicable. Stable layout frames and transition frames answer different questions.
  2. Extract — ffmpeg -ss <t> -frames:v 1 -update 1 -q:v 2 per timestamp.
  3. Inspect — Use the available image-viewing tool so the model sees the actual frame; Bash(cat / tail) will NOT show the image.
  4. Classify — for every finding, assign one of the four category numbers.
  5. Report — present findings with timestamps and visible evidence. A review request does not authorize edits; if fixes were requested, implement clear defects and ask only about choices that materially change the intended design.
  6. When fixes are in scope, re-render and inspect the same timestamps. Compare before/after; check for collateral regressions in adjacent bits.

Workflow details in rules/audit-workflow.md and rules/reporting.md.

File map

FilePurpose
rules/audit-workflow.mdffmpeg extraction + Read inspection + timestamp selection + iteration loop
rules/reporting.mdThe reporting format to the user (Bit / category / location), tone rules, before/after table format
rules/pixel-ratchet.mdExisting pixelmatch-py frame comparison and evidence limits
rules/typography-drift-gate.mduharfbuzz typography gate reference; verify current code before using historical snippets
references/defect-catalogue.md12+ verified incidents (filewalk × 5, hero × 7) — location, symptom, fix
references/external-oss.mdYusuke710/manim-skill (4-phase loop), pytest-mpl (baseline UX), pixelmatch-py, uharfbuzz, frames-mcp — stars / last-push verified 2026-05-21

Quick reference — extract + inspect

Use the isolated extraction example in audit workflow, then inspect each PNG with the available image viewer. Preserve prior renders and audit frames. KO is audited separately because Korean text width differs from English — typographic and padding defects do not transfer 1:1.

Standing line of defence

The verify_hero_layout.py ratchet (covered in [[manim-scene-craft]]'s rules/layout-ratchet.md) catches text-vs-container geometry overflow at compile time. This skill's job is everything the geometric ratchet cannot see: arrow / label spatial relationships, typography drift inside the box, and animation timing. Steps 2 and 3 (uharfbuzz, pixelmatch) extend the compile-time ratchet so the model spends less time inspecting frames manually — see rules/typography-drift-gate.md and rules/pixel-ratchet.md.

Reviews

No reviews yet. Tell others what you thought.

Comments

No login needed. Share how it worked for you, ask a question, or note where it did not run.

No comments yet. Be the first to write one.

Comments are published right away. Do not include personal information such as email addresses or phone numbers. Up to 2 URLs.

0 / 1,000 characters

This form is protected by reCAPTCHA, and the Google Privacy Policy and Terms of Service apply.

More skills from this repository

agent-anti-pattern

Free

Audit agent runtimes and their tests for lifecycle inference, capability overload, context pollution, correlated verification, prompt-only control, false success, dead compatibility paths, test-only seams, and complexity without contribution. Use for whole-repository agent audits, slop or dead-code triage, test contribution reviews, large refactors, tool/context simplification, and cleanup requests that need proof before deleting code or tests.

The English summary is being prepared. Showing the original description.

mangowhoiscloud/geode15Updated Oct 16, 2026

agent-ops-debugging

Free

Diagnose agent operational failures through default-policy checks, multi-gap analysis, explicit service ownership, request-context lifetime, and degradation-versus-correctness checks.

The English summary is being prepared. Showing the original description.

mangowhoiscloud/geode15Updated Oct 16, 2026

agent-world-benchmark

Free

Align, run, audit, and publish GEODE benchmark comparisons against Agent-World arXiv 2604.18292v1 across MCP-Mark, BFCL V4, and tau2-bench. Use for Agent-World or AgentWorld comparison requests, cross-suite agent benchmark design, Tau2 comparator/profile diagnosis, paired thin-runtime-versus-GEODE experiments, mean-accuracy@8 replication, or benchmark comparability and artifact reviews.

The English summary is being prepared. Showing the original description.

mangowhoiscloud/geode15Updated Oct 16, 2026

anti-deception-checklist

Free

Verification checklist to prevent fake success. Detects test deletion/disabling, coverage regression, lint bypass, secret exposure. Triggered by "deception", "fake" ("가짜"), "fake success", "verification" ("검증"), "checklist" ("체크리스트"), "deletion detection" ("삭제 탐지"), "regression" keywords.

The English summary is being prepared. Showing the original description.

mangowhoiscloud/geode15Updated Oct 16, 2026

architecture-patterns

Free

Implement proven backend architecture patterns including Clean Architecture, Hexagonal Architecture, and Domain-Driven Design. Use when architecting complex backend systems or refactoring existing applications for better maintainability.

The English summary is being prepared. Showing the original description.

mangowhoiscloud/geode15Updated Oct 16, 2026

arxiv-digest

Free

Auto-search and summarize latest AI/agent papers.

The English summary is being prepared. Showing the original description.

mangowhoiscloud/geode15Updated Oct 16, 2026

See all skills by mangowhoiscloud

Report a problem with this skill