argos-pr-review
Skill661 starsChanged 7 days agoMITView on GitHub
Review Argos visual regression builds as one input to a pull request review. Use when a PR has an Argos build link, an Argos status check, or a bot comment pointing to an Argos build, and you need to decide whether the visual diffs match the developer's intent before approving.
In this repository9 other files it gives its agents.
- AGENTS.mdAGENTS.md
- .agents/skills/vercel-react-best-practices/AGENTS.mdAGENTS.md
- argos-cliSkill
- argos-uploadSkill
- frontend-designSkill
- lucide-style-iconsSkill
- security-auditSkill
- upgrade-depsSkill
- vercel-react-best-practicesSkill
The file
We split this skill into the rules it gives an agent
3 rules. About 0 are common, 0 are distinctive, and 3 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈0
Distinctive · estimate
This author's own choices. Worth a look.
3
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 17
Treat the Argos build as one input to the PR review, never the sole source of truth.
- Very specific: line 82
Use --draft to bundle comments into your pending review, then submit them together in the next step.
- Very specific: line 89
Add --project owner/project for build-number refs (not URLs).
≈ “Common” is an estimate: at least 1 in 5 of the 372,429 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
Argos PR Review
Argos is a visual testing platform: each CI build captures screenshots and
compares them to a baseline, producing per-snapshot diffs. A build with
changes-detected is waiting for a human (or you) to decide whether each change
is intentional, a regression, or a flaky capture.
Treat the Argos build as one input to the PR review, never the sole source of truth. Infer the intended UI change from the PR title, description, linked issue, and code diff first, then use Argos to confirm the rendered result matches.
Tooling & auth
Drive Argos through the argos CLI — load the argos-cli skill for the token
model and flags. In short: build get / build snapshots need a project token
(ARGOS_TOKEN/--token); submitting a review or comment needs a personal
access token (--token / argos login). Always use --json when parsing, and
never print token values. If no PAT is available, give the user your conclusion
and evidence instead of posting — the CLI can't submit the review.
Workflow
Inspect the build —
argos build get <ref> --json. Decide from status:Status Meaning / next step accepted/no-changesAlready approved / no visual diff — no review needed pending/progressNot ready — stop and report it can't be reviewed yet changes-detectedNeeds a decision — fetch snapshots rejected/error/aborted/expiredDon't approve until the cause is understood - Fetch what changed —
argos build snapshots <ref> --needs-review --json. For each diff inspecturl(diff mask),base.url(before),head.url(after), andhead.metadata, plus the flakiness signalstest.metricsand, on a change,change.occurrences/change.ignored(used in step 3). - Judge each diff against the inferred intent:
- Intentional — matches the code change and renders cleanly.
- Regression — broken layout/overlap, clipping, wrong state/theme/route, missing content, or a removed snapshot with no matching test removal.
Flaky — weigh two independent signals; when they agree, call it flaky with confidence:
- Test history —
test.metrics.flakiness(0 stable → 1 flaky) andchange.occurrences(how many times this exact diff has recurred over the metrics period). A high flakiness score or a recurring change is strong evidence the diff is environmental noise, not this PR's work;stability(builds without a change),consistency(do changes repeat identically), anduniqueChangesexplain why it scores that way.change.ignored: trueis already known-flaky and auto-approved — never read it as a regression. - This capture — a spinner/skeleton, async content not yet loaded,
mid-animation, drifting dynamic values,
head.metadata.test.retry > 0, or identicalscore/head.urlacross browsers (both captured the same transient state).retries(the configured budget) is not itself a signal.
Conversely, a stable test (
flakiness→0, highstability, a first-time change) that changed is more likely intentional or a real regression — don't dismiss it as flaky on the visuals alone. Tune the window withbuild snapshots --metrics-period <24h|3d|7d|30d|90d>(default7d).- Test history —
Comment on specific diffs — the highest-value output of an agent review. A binary approve/reject is cheap; specific, anchored feedback is what makes an agent review worth reading. For each problem diff, post a comment that names what's wrong and how to fix it, anchored to that snapshot:
argos comment create <ref> --token <pat> --diff <screenshotDiffId> \ --body "Loader still visible — capture runs before data loads. Wait for settled content (or mark the loader aria-busy)."<screenshotDiffId>is the diffidfrombuild snapshots --json.- Pin to a region with
--anchor-lines <from,to>or--anchor-point <x,y>(normalized 0–1); reply in a thread with--reply-to <commentId>. - Use
--draftto bundle comments into your pending review, then submit them together in the next step.
- Submit the review (or report if no PAT is available):
- Approve:
argos review create <ref> --token <pat> --event approve - Request changes:
argos review create <ref> --token <pat> --event reject --body "<summary>" - Neutral note only:
--event comment --body "<summary>" - Add
--project owner/projectfor build-number refs (not URLs). - Silence a confirmed recurring flake so it stops blocking future builds:
argos change ignore <change.id> --token <pat> --project owner/project(reverse withchange unignore). Only ignore flakes the metrics confirm — never to bypass a real diff; prefer a code fix (references/flaky-fixes.md) when one exists. - Lead with the inferred intent, the snapshots reviewed, and the evidence. In the PR, cite the build URL and affected snapshot names; for flakes, name the signal and recommend a fix.
- Approve:
References
- references/baseline.md — baseline selection and orphan-build semantics (load when the review depends on which baseline was used).
- references/flaky-fixes.md — concrete code fixes for
flaky captures (
aria-busy,data-visual-test, animation stabilization).
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.

