Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx agentmods add skills/jangles-byte/atelier/critiquenpx skills add jangles-byte/atelier --skill critiquegit clone --depth 1 https://github.com/jangles-byte/atelierWhat it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5 | $0.00123 | $0.00663 |
| Opus 5 | $0.00062 | $0.00331 |
| Sonnet 5 | $0.00025 | $0.00133 |
| Haiku 4.5 | $0.00012 | $0.00066 |
Grade A, and why
critique scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 2d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 49 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Critique
A critique is an ordered list of changes with exact values, produced by evaluating a real render against a fixed rubric. "Feels unbalanced" is not critique; "increase section gap 32px→64px so the pricing table separates from the hero" is.
Workflow
- Get a real render. Screenshot the running artifact (browser preview, simulator,
plot output, game capture) — never critique source code or imagination.
If anything moves, a still is not a render. Capture the motion and watch it:
Then interact with it directly — hover, open and close, mash the trigger — because interruption bugs only appear under abuse.../motion/scripts/capture-motion.py index.html --out review.gif --duration 1200 - Inventory — one paragraph of what is objectively on screen (elements, reading order as encountered, palette, faces in use). Forces looking before judging.
- Score the rubric — eight dimensions, 1–5 each, using references/rubric.md. Cite evidence per score.
- Write the change list — every scored weakness becomes a change in the exact format of references/change-format.md: ranked by impact, each with location, current value → new value, and the principle it serves. 3 changes minimum, 10 maximum; past 10, ship the top 10 and re-critique after.
- Apply and re-render. A critique that isn't applied and re-verified is a memo. Iterate until the weakest rubric dimension scores ≥ 4 or the user stops you.
Which reference to load
| Situation | Load |
|---|---|
| Scoring any render; the eight dimensions with anchors | references/rubric.md |
| Writing the output; worked example of a full critique | references/change-format.md |
Rules
- Judge against the design philosophy first (does it keep its own promises?), the rubric second. If no philosophy exists, write a one-line inferred intent and say so.
- Diagnose over prescribe when uncertain of intent — but never emit a finding without a proposed exact value. Uncertainty goes in a parenthetical, not in vagueness.
- Strengths get one sentence, findings get the rest. Flattery is not calibration.
- Accessibility findings (contrast, motion, hit targets) always rank above aesthetic findings, whatever their scores.
What ships with it
2 files beside SKILL.md in the same directory: the scripts, references and assets a skill reads on demand. Not counted in the per-session cost; read them before you install if any of them is executable.
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 2d ago First seen · 49 lines · 0 tokens per session scan A c2a7f108ac56
critique is a skill published in the GitHub repository jangles-byte/atelier (2 stars, last pushed 1mo ago), licensed MIT. It adds 123 tokens to every session and 663 once invoked, about $0.0006 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.
Other skills, from other repositories
sheleg-design
Use when deciding how something LOOKS or MOVES — cinematic landing pages and hero sections, particle/WebGL backgrounds, scroll-linked or scrubbed motion, layers that drift, dashboards, admin or internal tools, mobile screens, chat or agent interfaces, tokens, themes, palettes, typography and the Figma border. Triggers…
html-ppt-zhangzara-cobalt-grid
OpenDesign renewal + seat-expansion business case for a growing customer: realized value, usage proof, and the expansion ROI. Built as a decision-grade B2B sales deck for champion, finance approver.
html-ppt-zhangzara-mat
A margin-recovery diagnosis for a regional grocery chain — the governing thought, the driver tree, the priorities, and the roadmap. Built as a decision-grade consulting deck for client sponsor, steering committee.
html-ppt-zhangzara-peoples-platform
A public-transit funding proposal for a city council — ridership need, the plan, the risk controls, and the ask. Built as a decision-grade policy briefing deck for city council, public board.
html-ppt-zhangzara-long-table
OpenDesign's unit-economics and BYOK cost model: the assumptions, the sensitivity, and why it scales. Built as a decision-grade data & finance deck for CFO, investors.
hr-onboarding
A new-hire onboarding plan as a single page — first week schedule, buddy + manager intro, learning track, equipment checklist, and "you're set when…" outcomes. Use when the brief mentions "onboarding", "new hire", "first week plan", or "入职".