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/cavalry-collective/visual-stack/reviewnpx skills add Cavalry-Collective/visual-stack --skill reviewgit clone --depth 1 https://github.com/Cavalry-Collective/visual-stackWhat 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 | $0.00140 | $0.07958 |
| Opus 5 | $0.00070 | $0.03979 |
| Sonnet 5 | $0.00028 | $0.01592 |
| Haiku 4.5 | $0.00014 | $0.00796 |
Grade A, and why
review 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 — 478 lines — stays where its author put it; the contents beside it link to each section on GitHub.
0 · Host (do this first)
This skill is tool-agnostic. The loop is defined in
plugins/vstack/contracts/review-loop.md and the ops you need in
plugins/vstack/contracts/host.md. Never invent host-specific tool names in
this file — they live only in the Host adapter.
| You are running under… | Load adapter | Set |
|---|---|---|
| Codex | skills/review/hosts/codex.md |
VSTACK_HOST=codex (or --host codex on serve) |
| Grok Build / Grok CLI | skills/review/hosts/grok.md |
VSTACK_HOST=grok (or --host grok on serve) |
| Claude Code | skills/review/hosts/claude.md |
VSTACK_HOST=claude (default if unset) |
Adapters live in the hosts/ directory beside this SKILL.md. Do not read
plugins/vstack/host-profiles/<id>.json instead: that JSON is UI data with no
tool mapping.
Read the adapter before §3. Every background, watch_stream, stop,
share, and browser_capture step is fulfilled exactly as that file says.
A two-way review loop. The user comments on the screen; you apply the comments, ask about anything ambiguous, and publish the next version. Two things can go under it:
| what it is | what you change | |
|---|---|---|
| A page (§1–§3) | a wireframe you just generated, an exported screen, a prototype — any self-contained HTML file | the file |
| A UI that exists (§7) | an app on localhost, or a website on the internet: the real screens, real data, real states | the source code — or, for a site you don't own, a note about it |
requirements ──► page.html ──► review workspace ──► brief.md ─────┐
▲ or a live app (user comments) │
└──────── you apply it, reply, publish v(N+1) ◄──────────────┘
One review is one subject. A page review opens that file and nothing else — several screens means several files and several workspaces, not an invented project structure. A live review opens the whole app, because clicking through it is the point.
Which one? If the thing the user wants to talk about already runs, review it running (§7) — a wireframe of a screen that exists is a copy that drifts. Build a page when the screen does not exist yet, or when they want to change it without changing code.
What ships with it
16 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.
- assets/bundle-artifact.mjs 5.2 KB runs code
- assets/harvest-reference.js 5.7 KB runs code
- assets/review-server.mjs 91 KB runs code
- assets/screen-template.html 2.1 KB
- assets/workspace.html 263 KB
- hosts/claude.md 3.5 KB
- hosts/codex.md 4.6 KB
- hosts/grok.md 4.5 KB
- references/design-sources.md 4.3 KB
- references/workflow.md 17 KB
- tests/design-tokens.mjs 5.9 KB runs code
- tests/host-profiles.mjs 2.5 KB runs code
- tests/review-lifecycle.mjs 11 KB runs code
- tests/round-gate.mjs 9.4 KB runs code
- tests/update-check.mjs 3.8 KB runs code
- tests/workdir.mjs 3.2 KB runs code
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 · 478 lines · 140 tokens per session scan A f2ccd4f18d72
review is a skill published in the GitHub repository Cavalry-Collective/visual-stack (11 stars, last pushed 13d ago), licensed MIT. It adds 140 tokens to every session and 7,958 once invoked, about $0.0007 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 2026-08-30.
Other skills, from other repositories
ideogram-ultra
Build Ideogram 4 (Ideogram Ultra) txt2img and img2img workflows with the local open-weights model, dual conditional/unconditional models with DualModelGuider, Qwen3-VL text encoder, and structured JSON ("compositional deconstruction") prompts for strong text rendering and layout control.
model-compatibility
Model family compatibility matrix covering loaders, resolutions, samplers, CFG, VAE, ControlNet, and LoRA compatibility for SD 1.5, SDXL, Flux, SD3, and video models.
flux-txt2img
Build Flux txt2img workflows with Flux.1 Dev (SRPO), Flux 2 Klein 9B, Turbo LoRAs, FluxGuidance, and DualCLIPLoader patterns.
panel-node-pack-sync
Keep the ComfyUI sidebar panel node-pack (comfyui-agent-panel) in step with the orchestrator after comfyui-mcp updates. Use this whenever the orchestrator was just updated (installcomfyui (action:"selfupdate"), npm i -g comfyui-mcp, a new version in the ENVIRONMENT line), when a panel/bridge command fails in a way…
civitai
Discover Civitai models with the BUILT-IN downloadmodel action:"searchcivitai" and install/generate them locally. Find a checkpoint/LoRA/embedding on Civitai, download it into ComfyUI, and use its trigger words. Optionally pair the official Civitai MCP for community features (images browsing, posting, collections).
20-client-intake-brief-global
Use when an agency or freelancer takes on a NEW client and needs everything up front — an 11-section intake covering business overview, audience, current marketing, goals, budget, timeline, brand assets, approval process, and success criteria, written for the client to fill in and return. Trigger on 'client intake'…