本文へ移動
cccskills
無料GitHub で公開

gaia-image-production

Gaia Skill Tree image-generation and deterministic export harness for Ascension Overdrive v3. Use when generating, iterating, refining, cropping, color-grading, making transparent cutouts, compressing, upscaling, or validating production image assets. Always use image gen 2 / gpt-image-2 and never nano-banana.

インストール方法を見る

含まれるファイル(1)

  • SKILL.md10.5 KB

SKILL.md(原文)

インストールする前に、エージェントに与えられる指示の中身を確認できます。

Gaia Image Production Harness

Use this skill for generated images, image iteration, asset production, responsive crops, transparency, color grading, upscaling, motion posters, and final exports for Gaia Skill Tree. For Ascension Overdrive v3, the commissions document is the source of truth; do not invent assets outside it.

This skill is adapted from the Gaia Research image-production harness. Its Sharp-based ledger helpers remain in the adjacent Gaia Research repository and are not assumed to exist here. Gaia Skill Tree uses the deterministic local toolchain documented below.

Hard rules

  • Always use image gen 2 / gpt-image-2 for generative work.
  • Never use nano-banana, nano-banana-2, or a cheap/fast image model for Gaia production assets.
  • Preserve every supplied or generated master outside Git. Never overwrite an incoming master.
  • Only production WebP raster derivatives enter this repository. Keep PNG, PSD, AI, Figma, and other raster masters in the local master archive.
  • SVG and motion formats enter Git only where the commissions document explicitly requires them, notably Asset H's primary SVG and Asset E's loops.
  • Do not regenerate an accepted composition when deterministic editing, compositing, color grading, or export can satisfy the brief.
  • Do not modify or replace v2 assets marked REUSED-V2.
  • Use the aov3- filename prefix for every new Ascension v3 asset.

Canonical Ascension v3 paths

  • Commission source of truth: founder/handovers/design-v6.1.1-ascension-overdrive-commissions-v3.md
  • Section shape: founder/handovers/design-v6.1.1-ascension-overdrive-shape-v3.md
  • Existing v2 implementation reference: docs/css/ascension-overdrive-v2.css and docs/js/ascension-overdrive-v2.js
  • Served production assets: docs/assets/ascension-overdrive/
  • Local master root, outside Git: /Users/marcotiongson/Documents/gaia-asset-masters/ascension-overdrive-v3/
  • Immutable supplied originals: /Users/marcotiongson/Documents/gaia-asset-masters/ascension-overdrive-v3/incoming/
  • Working masters and iterations: /Users/marcotiongson/Documents/gaia-asset-masters/ascension-overdrive-v3/workbench/
  • Approved local masters: /Users/marcotiongson/Documents/gaia-asset-masters/ascension-overdrive-v3/approved/
  • Local review sheets: /Users/marcotiongson/Documents/gaia-asset-masters/ascension-overdrive-v3/review/

Ascension v3 visual register

The design is a deliberate dichotomy, not one visual system recolored twice:

  • Suite is warm, ceremonial, ordered, engraved, architectural, and closed.
  • Unique is cold, broken, void-filled, impossible, and open.
  • Shared substrate is celestial copperplate cartography, never a modern nebula or fantasy map.
  • Suite stamps ascend from natural-history specimen plates at 1–3 stars to Beaux-Arts ceremonial architecture at 4–6 stars.
  • Unique geometry must break structurally. It must not be Suite ornament on a dark background.
  • Asset H is one binary gold thread: one trunk, exactly two branches, with negative space reserved for card overlays.

Use the OKLCH targets in the commissions document as the authoritative palette:

  • Suite cream: oklch(0.92 0.025 85)
  • Suite gold: oklch(0.78 0.14 82)
  • Suite brass: oklch(0.65 0.10 72)
  • Suite ink: oklch(0.25 0.02 75)
  • Unique void: oklch(0.13 0.015 260)
  • Unique singularity: oklch(0.96 0.01 240)
  • Unique fracture: oklch(0.55 0.03 245)
  • Unique glitch: oklch(0.72 0.16 12)
  • Substrate ground: oklch(0.22 0.015 85)
  • Substrate ink: oklch(0.55 0.03 75)

Standard workflow

1. Read the brief and references

Read the relevant asset section in the commissions document, its failure test, the v3 shape, and the closest accepted v2 production references. Treat exact dimensions, alpha requirements, mobile compositions, size ceilings, named SVG paths, and reuse/deprecation status as acceptance criteria.

2. Intake originals without mutation

Copy supplied files into incoming/ with a descriptive -concept-original name. Record the original filename, dimensions, byte size, and SHA-256 in the directory README. Verify source and copy hashes match. Never optimize, resize, strip metadata, or color-grade an incoming original in place.

3. Plan the batch

Before generation or editing, record:

  • asset ID and purpose;
  • target scene and rail;
  • desktop, mobile, card, and badge dimensions required by the spec;
  • alpha or opaque-ground requirement;
  • reference masters that must be preserved;
  • target WebP size ceiling;
  • visual acceptance criteria and the applicable numbered failure tests.

4. Generate only what the brief requires

Use image gen 2 for missing compositions or targeted visual refinement. Use the selected image as the reference for chained iteration. Ask for only the desired change, retain versioned masters, and keep quiet space or silhouettes required by the scene. Do not ask the model to rasterize UI copy.

For concept exploration, vary one meaningful axis per branch, such as geometry, silhouette, negative-space distribution, or material register. Select one candidate before expensive upscale and export work.

5. Prefer deterministic edits after selection

Once a composition is accepted, use deterministic tools for crop, resize, color-grade, alpha cleanup, metadata stripping, and WebP compression. Avoid generative edits for operations that should be reproducible.

The verified local stack is:

  • Pillow 12 with libwebp 1.6 for scripted pixel operations;
  • cwebp 1.6 for final WebP encoding, alpha control, crop, resize, and metadata stripping;
  • ffmpeg and ffprobe 8.1 for Asset E loops and poster extraction;
  • sips for quick metadata inspection;
  • SHA-256 via shasum -a 256.

ImageMagick and a project-local Sharp installation are not assumed available. Do not add them merely to complete routine production work.

6. Export WebP deliberately

For an opaque plate or substrate:

cwebp -preset picture -m 6 -q 78 -sharp_yuv -metadata none \
  input.png -o output.webp

For an alpha-bearing stamp or overlay:

cwebp -preset picture -m 6 -q 82 -alpha_q 100 -exact -sharp_yuv \
  -metadata none input.png -o output.webp

For a deterministic crop and resize before encoding:

cwebp -preset picture -m 6 -q 82 -alpha_q 100 -exact -sharp_yuv \
  -metadata none -crop X Y WIDTH HEIGHT -resize OUT_WIDTH OUT_HEIGHT \
  input.png -o output.webp

Tune quality downward only as far as the asset's size ceiling requires. Inspect engraved hairlines, alpha edges, dark gradients, and singularity highlights at 100% after every compression change. Do not force card and badge variants from the hero master when the spec requires hand-composed small-scale artwork.

7. Prepare transparency and chroma-key cutouts

Perform keying on a versioned workbench copy. Preserve RGB under transparent pixels when edge compositing benefits from it. Produce a diagnostic mask and a checkerboard preview, then inspect the result over both the substrate ground and the Unique void. Reject white halos, dark matte fringes, pinholes, and clipped engraved edges.

The Gaia Research prep-cutout.ts helper is a useful implementation reference, but it resolves paths against Gaia Research and must not be run against this repository without a deliberate port. Pillow is the default local mechanism until such a port is separately commissioned.

8. Color-grade Asset G before commissioning new haze

Create Suite-warm and Unique-cold grades from the preserved v2 haze source. Keep the alpha field and luminance structure stable while changing hue and temperature. Launch a fresh illustration only if one of the explicit Asset G escalation triggers in the commissions document fails during compositing.

9. Handle Asset E motion deterministically

Keep master loops outside Git. Deliver 24 or 30 fps, silent, seamless 6–12 second loops. Use H.264 MP4 and VP9 WebM only if the spec-authorized motion formats are being committed. Extract a first-frame-parity WebP poster for the reduced-motion path.

Example poster extraction:

ffmpeg -i input.mp4 -vf 'select=eq(n\,0)' -frames:v 1 poster.png
cwebp -preset picture -m 6 -q 82 -sharp_yuv -metadata none \
  poster.png -o poster.webp

Inspect loop duration, frame rate, dimensions, audio absence, and codec with:

ffprobe -v error -show_entries stream=codec_name,width,height,r_frame_rate \
  -show_entries format=duration,size -of json input.webm

10. Review in contact sheets and composites

Review each related set at equal display size on one sheet. For C and D, place 4–6 star Suite and Unique stamps side by side so the dichotomy can be judged without labels. Also composite alpha assets over both rail grounds. Review hero, card, badge, desktop, and 375px mobile presentations independently.

11. Validate before promotion

For every production derivative, verify:

  • exact pixel dimensions;
  • actual file format, not only extension;
  • alpha presence where required;
  • byte size within the per-asset ceiling;
  • filename follows aov3-<asset>-<rank>-<state>.webp;
  • no collision with v2 filenames;
  • clean compositing on both required grounds;
  • source and derivative hashes recorded in the local master archive;
  • only the reviewed WebP derivative is staged as a raster asset.

Useful inspection commands:

sips -g pixelWidth -g pixelHeight -g format -g hasAlpha asset.webp
du -h asset.webp
shasum -a 256 asset.webp
file asset.webp

Run the repository's design/asset checks that apply to the changed files. Do not use unrelated full-suite failures to change the visual brief.

Ascension v3 delivery priorities

Follow the commission cadence unless Marcus changes it:

  1. Asset H, Y-Fork gold-thread illustration.
  2. Asset B v3, astrolabe substrate.
  3. Assets C v3 and D v3 as one paired stamp calibration pass.
  4. Asset I, Unique Impossible terminal.
  5. Asset F v3 Suite plates and Asset G color grades.
  6. Asset E v3 motion loops last.

Final handoff checklist

Report:

  • immutable source path and checksum;
  • approved local master path;
  • production WebP path and dimensions;
  • derivative paths and file sizes;
  • whether alpha/composite tests passed;
  • which commission failure tests passed;
  • which design checks ran;
  • any remaining manual visual-review risk.

レビュー

まだレビューはありません。使ってみた感想をお寄せください。

同じリポジトリのスキル

概要と使いどころ

ci-churn

無料

Measure CI churn cost for a pull request: how many commits and how many CI-compute seconds were burned on avoidable "fix the CI" push rounds. Use when you want to understand the efficiency cost of a PR's iteration cycle, get actionable pre-push check suggestions, or include a churn summary in a pipeline report. Triggers: "ci churn", "how much CI did we burn", "avoidable commits", "ci efficiency", "wasted CI time", "churn report", "/ci-churn". Called automatically by /fp-drift at pipeline close.

日本語の概要は準備中です。原文の説明を表示しています。

gaia-research/gaia-skill-tree232026年10月13日 更新

Produce the rendered evidence a human-gated frontend PR owes its reviewer: capture each changed surface (plus the before state for anything modified, plus mobile width), then generate one self-contained local HTML page the founder can open and review without checking anything out. Use when a PR is gated by the "Frontend changes are HUMAN-GATED" section of CLAUDE.md, or when the user says "/design-gate", "screenshot evidence for this PR", "prep this for founder review", "build the design evidence page", "what does this look like before and after". The page comes from a template script, not hand-written markup. Output is session-local and never committed. Do NOT use for non-gated changes: typo/grammar/factual fixes in an unchanged block, dead link targets, alt/ARIA fixes with no visual change, cache-bust bumps, or refactors with no rendered diff.

日本語の概要は準備中です。原文の説明を表示しています。

gaia-research/gaia-skill-tree232026年10月13日 更新

One-fix-at-a-time, approval-gated design fix recovery. Given a checklist of fixes (handover doc, CHECKLIST.md, or inline list), applies each individually, presents a diff, and waits for an explicit APPROVE / REJECT / NO CHANGES decision before committing. Trigger phrases: /design-iteration, "go through the checklist one by one", "fix these one at a time", "bucket recovery", "design fix pass".

日本語の概要は準備中です。原文の説明を表示しています。

gaia-research/gaia-skill-tree232026年10月13日 更新

Playbook for correcting a named skill's star level to match its computed Trust Magnitude grade, with an independent post-calibration verification pass and a CLI-logged timeline event (never hand-written). Use when: recalibrating a skill's level, fixing a stale or drifted star rating, running a batch of `gaia dev calibrate` corrections, or confirming a past calibration actually landed. Triggers: "calibrate <skill>", "recalibrate the trust magnitude backlog", "promote/demote <skill> to N stars", "why is <skill> still N stars", "verify a calibration actually landed", "Hall of Heroes ranking is wrong".

日本語の概要は準備中です。原文の説明を表示しています。

gaia-research/gaia-skill-tree232026年10月13日 更新

Close out a completed EPIC / sprint end-to-end: author a fact-verified meta-post announcing what shipped, archive the EPIC's handovers, snapshot orchestrator memory, gate the integration→main merge on CI-green + explicit human go, drive the version bump (merge-commit conventions that auto-sync classifies), and verify the release reached the registry/PyPI. Use when the user says "close out EPIC <n>", "wrap up the sprint", "ship the meta shift", "/epic-close", or asks to finalize an integration branch that is ready to merge to main. Generalizes the EPIC 1002 (Yggdrasil II) close-out. This skill orchestrates — it authors founder docs directly (superadmin) and delegates code/regen to workers; it NEVER merges to main without the human gate.

日本語の概要は準備中です。原文の説明を表示しています。

gaia-research/gaia-skill-tree232026年10月13日 更新

Run this skill for Phase 3 of the evidence verification pipeline — the adversarial audit step. Use when you need to check the evidence data lake for bad data before ingestion: dead links, wrong URL formats (tree/ vs blob/), subjective wording ("elite", "high-quality"), stale migration notes, benchmark catalog misuse/vendor-claim leakage, or skills whose star evidence conflicts with classified evidence level. Triggers on phrases like: "audit the data lake", "adversarial check", "ev-adversarial-audit", "check for noise in evidence", "flag bad evidence", "run the audit phase", "quality check the by-type files", or any reference to Phase 3 of the pipeline.

日本語の概要は準備中です。原文の説明を表示しています。

gaia-research/gaia-skill-tree232026年10月13日 更新

gaia-research のスキルをすべて見る

このスキルの問題を報告する