Borrowing it
Nothing to install: this file belongs to cfpperche/picode. Take a copy, put it at the same path in your own repository, and replace the rules that are about this project with yours.
curl -O https://raw.githubusercontent.com/cfpperche/picode/main/.pi/skills/visual-review/SKILL.mdgit clone --depth 1 https://github.com/cfpperche/picodeWrote 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/cfpperche/picode/visual-review)<a href="https://agentmods.dev/skills/cfpperche/picode/visual-review"><img src="https://agentmods.dev/badge/skills/cfpperche/picode/visual-review/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/cfpperche/picode/visual-review"><img src="https://agentmods.dev/badge/skills/cfpperche/picode/visual-review.svg" alt="Reviewed on agentmods" width="80" 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 | $0.00048 | $0.00862 |
| Opus 5.5 | $0.00019 | $0.00345 |
| Sonnet 5.5 | $0.00010 | $0.00172 |
| Haiku 4.5 | $0.00005 | $0.00086 |
Grade A, and why
visual-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 1mo 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 — 94 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Visual review
Code review cannot see pixels. eval JSON cannot see pixels. This skill
is the only valid visual verdict.
Instant FAIL (do not commit)
Any one of these on a captured screenshot or geometry audit:
- Clipped overlay — menu, popover, toast, slash list cut by the viewport
- Unreadable items — labels truncated to meaninglessness, date wrapping into sludge
- Occlusion — a control covered so it cannot be used
- Double scrollbar on one surface
- Dead hover — primary action paints the same color as its parent
- Geometry clip —
overlayAuditreportsok: false - Uneven row —
[data-align-row]children differ in height or top by more than 1px (toolbar/filter lines). Untagged rows are invisible to the auditor — mark them.
If you noticed a defect in reasoning and still said done, that is a contract violation (AGENTS.md honesty). Fix it or emit FAIL. Never ship it.
States (list / manager pages)
Happy-path-only is FAIL. Capture and read at least:
- empty (zero items)
- blocked (missing dependency / cannot act)
- one overlay if the flow has one
Where the evidence lives (ADR-0086)
Screenshots you read stay in var/screenshots/ (gitignored). They are not
committed: docs/screenshots/ is frozen history. The audit trail is the
visual-review: line in your reply and in docs/handoff/<date>-<branch>.md.
The loop
- Serve a scratch instance of the rebuilt binary
(
scripts/qa-scratch.sh start <name>), never production. - Act with
agent-browser(open the surface, click the control). - Measure after every overlay opens:
agent-browser eval 'JSON.stringify(window.__picodeOverlayAudit())'
ok: false → FAIL. Do not proceed to a verbal PASS.
- Capture a full-viewport PNG (
agent-browser screenshot). - Read the PNG with the
readtool. Zoom mentally: edges, overflow, contrast, hierarchy. - Answer the card (required in the reply, before "done"):
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.
- 1mo ago First seen · 94 lines · 48 tokens per session scan A 6d1d76958bd4
visual-review is a skill published in the GitHub repository cfpperche/picode (13 stars, last pushed 1mo ago), licensed Apache-2.0. It adds 48 tokens to every session and 862 once invoked, about $0.0002 per session on Opus 5.5. 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 2026-09-12.
Other skills, from other repositories
verify
Exercise the real app/API/CLI and collect observable evidence; tests alone do not count as end-to-end verification.
webapp-testing
Start/reuse a local app, wait for readiness, inspect rendered state/console/network, act from observed selectors, and verify with evidence.
site-preship-review
Use before publishing or relaunching any website — six parallel single-dimension reviewers (design, mobile, states, user flows, launch prep, security & infra) that render and look at every page, then one or two fix passes. Includes launch-readiness and rendered-page checks.
design-review-workflow
Enforces one structured end-of-build design/QA pass — never per-task — on the release build: a screenshot sweep matrix (every screen × light/dark × LTR+RTL × largest text scale × reduce-motion) plus an on-device pass on real cheap target hardware, findings graded BLOCKER/FIX/NOTE with every accessibility-floor…
ux-grill
Use when a running web app's UX has to be audited reproducibly rather than by feel — a deterministic mechanical pass over routes and viewports (axe, target size, horizontal overflow, page title, scripted journeys) followed by a screenshot-grounded interrogation of the operator, journey by journey, with two persona…
factory-render-verify
Render-and-measure receipts for any HTML page your factory builds — the render half of the design quality gate. Engineer runs it to screenshot every screen size and MEASURE what a source read or a single screenshot only guesses at: horizontal overflow, computed type sizes, tap-target sizes, safe-area presence, mono…