reviewing-visual-builds
Skill0 starsChanged 9 days agoMITView on GitHub
Reviews a SnapVisor visual regression build: shows what changed in the screenshot diffs for a branch or commit and approves or rejects the build after the user confirms. Use when the user asks 'what changed in the latest build', 'show the screenshot diffs', 'review the visual changes on this branch', 'approve build 42, the nav restyle is intentional', 'reject this build', or asks about visual testing or UI regression results.
In this repository66 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- adding-uptime-monitorsSkill
- analyzing-page-performanceSkill
- booking-meetingsSkill
- building-onboarding-workflowsSkill
- building-study-guidesSkill
- capturing-tasksSkill
- capturing-web-pagesSkill
- chasing-unpaid-invoicesSkill
- checking-domain-connectionsSkill
- checking-email-deliverabilitySkill
- checking-notification-deliverySkill
- checking-uptime-statusSkill
- cleaning-up-bookkeepingSkill
- connecting-customer-domainsSkill
- discussing-buildsSkill
- drafting-invoicesSkill
- drafting-social-postsSkill
- editing-demosSkill
- editing-page-draftsSkill
- extracting-document-textSkill
- finding-meeting-timesSkill
- handling-flaky-changesSkill
- investigating-tasksSkill
- making-demo-videosSkill
- managing-bookingsSkill
- managing-content-calendarSkill
- managing-demosSkill
- managing-voice-librarySkill
- planning-revisionSkill
- preflighting-domainsSkill
- publishing-pagesSkill
- reading-financial-reportsSkill
- reading-text-aloudSkill
- reading-web-pagesSkill
- reporting-snapvisor-usageSkill
- researching-course-materialSkill
- researching-summary-librarySkill
- responding-to-incidentsSkill
- reviewing-email-performanceSkill
- reviewing-notification-workflowsSkill
- reviewing-social-performanceSkill
- reviewing-workloadSkill
- searching-getitdone-api-docsSkill
- sending-email-campaignsSkill
- sending-notificationsSkill
- setup-bioflowSkill
- setup-calySkill
- setup-demofySkill
- setup-dodomainSkill
- setup-getitdoneSkill
- setup-notiflySkill
- setup-postifySkill
- setup-sendlySkill
- setup-shortySkill
- setup-snapvisorSkill
- setup-superbooksSkill
- setup-unotesSkill
- setup-upapiSkill
- setup-uptimelySkill
- The first 60 shown.
The file
We split this skill into the rules it gives an agent
7 rules. About 0 are common, 3 are distinctive, and 4 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈3
Distinctive · estimate
This author's own choices. Worth a look.
4
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 16
Operations listed by name: call them directly, for example snapvisor:listBuilds.
- Very specific: line 27
If the connection lists the operations by name instead (the per-tool surface), call them directly without the external_ prefix.
- Very specific: line 41
If you do not know the account and project, run one program: external_getMe({}), then external_listProjects for the account the user means.
- Distinctive: line 45
If the user wants a decision, restate it ("Approve build 42 in project web.
- Very specific: line 46
Call external_createReview with event: "APPROVE" or event: "REJECT" and the user's reason as body.
- Distinctive: line 63
Never file a review without the user's explicit confirmation in this conversation, and never approve or reject several builds in a loop.
- Distinctive: line 64
Always show the diffs before asking for a decision.
≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
<!-- Generated from plugins/snapvisor/skills/reviewing-visual-builds/SKILL.md by scripts/sync-skills.mjs. Edit the plugin copy, then run the script. -->Review a SnapVisor build
Walk the user through the visual changes SnapVisor detected in a build, then approve or reject it on their say-so. An approval is a real decision for the whole team: the build is accepted and its changed screenshots become the new baseline.
Connecting and calling the tools
Every operation in this skill comes from the snapvisor MCP server, the SnapVisor connector. The server defaults to Code Mode, so the operations below are written as the external_* functions you call inside snapvisor:execute_typescript. On a server set to the per-tool surface, the same operation is listed as snapvisor:<name> without the external_ prefix, for example snapvisor:listBuilds.
- Only
snapvisor:search_toolsandsnapvisor:execute_typescriptlisted (Code Mode, the default): callsnapvisor:search_toolsfirst, then call each tool insidesnapvisor:execute_typescriptasexternal_<tool>(for exampleexternal_listBuilds); the program mustreturnits result. Scopes, confirmations and gates are identical on both surfaces. - Operations listed by name: call them directly, for example
snapvisor:listBuilds. - "Tool not found": switch to the Code Mode route instead of retrying the direct call. If
snapvisor:search_toolsdoes not return the tool either, its OAuth scope was not granted; the user reconnects SnapVisor and approves that scope. - No
snapvisortools at all: SnapVisor is not connected yet. Follow thesetup-snapvisorskill.
Tools you will use
The snapvisor MCP server has two tools. Everything below goes through them.
snapvisor:search_tools: find the operations you need and their exact input shape. Call it before your firstsnapvisor:execute_typescript, for example with the queriesbuild,diff,review.snapvisor:execute_typescript: run a short program that calls theexternal_*functionssnapvisor:search_toolsdeclared. They are already in scope; never import or redeclare them. The program mustreturnits result.
If the connection lists the operations by name instead (the per-tool surface), call them directly without the external_ prefix.
Operations this skill uses (confirm names and inputs with snapvisor:search_tools):
external_getMe: the signed-in user and the accounts (withslug) this connection can reach.external_listProjects: projects in an account.external_listBuilds: builds of a project, newest first; filter by branch (head), commit (headSha) orsearch.external_getBuild: one build with status, conclusion, stats, branch and commit.external_listBuildDiffs: one entry per screenshot withname,status(added,changed,removed,failure,ignoredand more), and base and head image links. PassneedsReviewto see only what needs a decision.external_listReviews: reviews already filed on the build.external_createReview: file a review witheventAPPROVEorREJECTand an optionalbody.
Workflow
- If you do not know the account and project, run one program:
external_getMe({}), thenexternal_listProjectsfor the account the user means. Ask once if more than one project fits. - Find the build. With no build number given, take the newest from
external_listBuilds(or the newest on the branch the user named). - In one
snapvisor:execute_typescriptcall, fetchexternal_getBuild,external_listBuildDiffsandexternal_listReviewstogether withawait Promise.all([...])and return a compact summary. - Present the build number, branch, commit and status, then the changed screenshots grouped by status (changed, added, removed, failed). Give each its before and after links. Mention reviews already on the build.
- If the user wants a decision, restate it ("Approve build 42 in project web; 6 changed screenshots become the baseline") and wait for a clear yes.
- Call
external_createReviewwithevent: "APPROVE"orevent: "REJECT"and the user's reason asbody. Report the new review, and confirm the build's status withexternal_getBuild.
Code Mode pattern
const [build, diffs, reviews] = await Promise.all([
external_getBuild({ owner, project, buildNumber }),
external_listBuildDiffs({ owner, project, buildNumber, needsReview: "true" }),
external_listReviews({ owner, project, buildNumber }),
]);
return { build, diffs: diffs.results.map((d) => ({ name: d.name, status: d.status })), reviews };Wrap a call in try { ... } catch (e) { return String(e) } when it may fail. Errors read <CODE>: <details>; relay the code and details instead of guessing.
Rules
- Never file a review without the user's explicit confirmation in this conversation, and never approve or reject several builds in a loop. One build, one confirmation.
- Always show the diffs before asking for a decision.
- Image links are signed and expire after about an hour. Hand them over; do not store or archive them.
- A 403 saying MCP access requires a paid plan means the account is on the Free plan. Say so plainly and do not retry.
- SnapVisor compares screenshots the user's CI uploaded. Do not offer to capture, upload or edit a screenshot; no operation on this surface can.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

