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/metraton/gaia/visual-verifynpx skills add metraton/gaia --skill visual-verifygit clone --depth 1 https://github.com/metraton/gaiaWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/metraton/gaia/visual-verify)<a href="https://agentmods.dev/skills/metraton/gaia/visual-verify"><img src="https://agentmods.dev/badge/skills/metraton/gaia/visual-verify.svg" alt="Measured on agentmods" height="20"></a>What 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 5.1 | $0.00066 | $0.01372 |
| Opus 5 | $0.00033 | $0.00686 |
| Sonnet 5 | $0.00013 | $0.00274 |
| Haiku 4.5 | $0.00007 | $0.00137 |
Grade A, and why
visual-verify 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 5d 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 — 105 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Visual Verify
Visual verification is the discipline of confirming a visual output by rendering it and looking at the pixels, not by asserting the markup "should" look right. When work produces something a person would see -- a page, a UI, a diagram, a slide -- the honest check is a screenshot the agent then reads with its own image-reading tool. An agent that ships visual output without looking at it has not verified; it has hoped.
This skill teaches the disposition and hands you a reference implementation
(scripts/screenshot.cjs) as support, not as a spec. It is generic -- any HTML
or URL, any project, inside Gaia or out. For the response contract and the
verification block this feeds, see agent-protocol.
Core principle
Three judgments shape a good visual check. Reason through each; do not follow a fixed recipe.
- Look, and look where it breaks. Markup that parses is not layout that works. Text clips, boxes collide, a column that is fine at 1440px overflows at 380px, a palette that reads in light mode fails in dark. So render across a spread of widths (desktop down to narrow mobile) AND across the themes the output supports (light/dark) -- one viewport in one theme is not a verification of a responsive, themed surface. Reading the images is the verification; the script exiting 0 is not.
- Find the browser where it lives; obtain it only if it is truly absent.
Playwright caches its browsers in a location that varies by OS and by the
PLAYWRIGHT_BROWSERS_PATHoverride -- do not assume one fixed path. Locate the browser dynamically (resolve the highestchromium-*build present) and launch it via an explicitexecutablePath, so you use what is already on disk. If nothing is present anywhere, obtaining a browser is a legitimate, deliberate one-time step -- a real install you run with approval. The friction to avoid is not installing per se; it is letting a tool silently fetch a mismatched revision when a perfectly usable browser already sits in the cache. - Put the captures where their context wants them. The output location is a
decision, not a fixed folder. Read the context and choose:
- Backing a brief (inside Gaia): write them where that brief's evidence lives, so the capture becomes part of the audit trail.
- Generic / outside a brief: a temporary directory -- it keeps the workspace clean and is not a meaningful mutation -- or the location the user or context names. The reference script takes the out-dir as an argument precisely so this decision stays with you.
What ships with it
1 file 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.
- 5d ago First seen · 105 lines · 66 tokens per session scan A fef83c024585
visual-verify is a skill published in the GitHub repository metraton/gaia (3 stars, last pushed yesterday), licensed MIT. It adds 66 tokens to every session and 1,372 once invoked, about $0.0003 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
review-team
6つの専門レビュアーロールを並列実行し、consensusLevel（複数ロールの合意度）と Tech Lead レポート（top3指摘・blindSpots・consensusSummary）で結果を統合する マルチエージェントレビュー entry skill。 Parallel multi-role review with consensus scoring (consensusLevel) and Tech Lead report. Use when a major release needs exhaustive multi-angle review, or when a single-perspective review…
writing-plans
Use when you have a spec or requirements for a multi-step task, before touching code - produces a plan this plugin can parse, phase, and dispatch to a fleet.
mode
Adopt the lead role for this session: plan work, delegate to executors via relay, review reports, never implement large work directly. Invoke with /relay:mode.
verify
Machine-check an executor's report against its staged reality — TL;DR block well-formed, claimed files actually staged, declared counts cross-checked — stamping MALFORMED / MISMATCH / INCONCLUSIVE / COUNTS-MATCH. Invoke with /relay:verify, or when asked "check that report", "did it really do what it says", "verify X's…
grok-cli-runtime
Internal helper contract for calling the grok-companion runtime from Claude Code.
auto
Flip this lead session's autonomous posture: proceed by default on routine in-plan steps instead of asking, or go back to waiting. Invoke with /relay:auto, or when asked to "go autonomous", "stop asking me every time", "just proceed", "turn auto off", "what posture am I in".