メニュー
Audit announcement copy for false, unverifiable, or fabricated claims. TRIGGER when the user asks to check, audit, or honest-check any copy, or when writing or reviewing any announce.md file.
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
SOC 職業分類に基づく
Classify raw DCC forum posts into plugin/addon/HDA opportunities using Claude Haiku agents. Reads the latest raw posts file (or a specified path), classifies each post with Haiku, clusters kept entries by underlying need, and writes a deduplicated opportunities JSON.
Build and document a new Blender addon plugin end-to-end. TRIGGER when: code imports `bpy`, or user asks to create, build, or scaffold a new Blender plugin/addon. DO NOT TRIGGER when: user asks to edit an existing plugin or write a tutorial.
Write a step-by-step tutorial for a Funkworks DCC plugin. Use when a plugin exists at plugins/<dcc>/src/<name>.py and needs a tutorial written for plugins/<dcc>/docs/tutorials/<name>.md. TRIGGER when the user asks to write, create, or draft a tutorial for a plugin.
Design-direction advisor + 5-dimensional design review for visual work. Pick a design philosophy from a library of 20 named styles (Pentagram, Field.io, Kenya Hara, Sagmeister, etc.) before designing, or score a finished design on philosophy alignment / visual hierarchy / craft quality / functionality / originality (each 0-10) with a Keep / Fix / Quick-Win list. Includes an anti-AI-slop checklist for resisting generic AI default aesthetics. Derived from alchaincyf/huashu-design (Personal Use License — see LICENSE + ATTRIBUTION.md).
Use when working with color naming, color theory, color spaces, color definitions, or any task involving color knowledge - palettes, ramps, gradients, conversions, accessibility, perceptual matching, pigment mixing, print-vs-screen color, CSS color syntax, or historical color terminology. Use this skill whenever the user is choosing, comparing, generating, naming, converting, or explaining colors, even if they do not explicitly ask for "color theory."
Create hand-drawn Excalidraw diagrams, flows, explainers, wireframes, and page mockups. Default to monochrome sketch output; allow restrained color only for page mockups when the user explicitly wants webpage-like fidelity.
| name | honest-copy |
| description | Audit announcement copy for false, unverifiable, or fabricated claims. TRIGGER when the user asks to check, audit, or honest-check any copy, or when writing or reviewing any announce.md file. |
| argument-hint | [path/to/announce.md or inline copy] |
| allowed-tools | Read, Glob, Grep, Bash |
Audit the announcement copy at $ARGUMENTS for honesty.
Read the file. Then go line by line through every claim and flag anything that fails one of these tests:
1. First-person experience claims Any sentence starting with "I", "I've", "Every time I", "I got tired of", etc. Ask: is this verifiably true based on the git history, issue tracker, or something the user has explicitly stated? If not, flag it.
Bad: "The Houdini COP I got tired of rebuilding from scratch" Bad: "Every time I needed to fit an image, I'd rebuild the same node chain" OK: "Houdini's Resample COP only does Stretch" ← verifiable fact about the software
2. Ordinal and superlative claims
"First release", "first Houdini release", "most complete", "the only", etc.
Run gh release list to verify any ordinal claim. Flag unsupported superlatives.
3. Implied repeated personal experience Narrative framing like "every time", "I kept having to", "I always ended up" implies something happened repeatedly. This is fabricated unless the user has said so. Rephrase as a description of the problem in second or third person instead.
4. Emotional/rhetorical filler "which feels right", "that's the way it should be", "finally", "at last" — these add sentiment without adding information. Flag and suggest cutting.
Also reduce extra metaphors. Stock figures like "papercut", "tedious dance", "shuffle" hide the actual mechanism — describe the literal behavior instead. ("Friction" and "pain point" are fine; they're plain industry terms.)
List each flagged item as:
LINE: [quote the sentence]
PROBLEM: [which test it fails and why]
FIX: [a replacement that says the same thing honestly, or "cut it"]
If nothing is flagged, say so explicitly: "No issues found."
After the audit, ask the user which fixes to apply, then edit the file.
The publish skill calls for a factual audit in Step 6. This skill is the implementation of that audit. Run it on every announce.md before outputting final copy.