Ui evidence
Local before/after UI evidence for agent-driven changes. Capture screenshots, build comparisons, and review what actually changed.
npx -y skills add 0xBrewing/ui-evidence --skill ui-evidenceAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
One thing to look at
- 5 stars5 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
What its author says it does
Copied from the file, not written here
Use when an agent needs to bootstrap ui-evidence in a repo, then discover, configure, or run before/after UI screenshot capture, current UI snapshot capture, local HTML review generation, or git-baseline UI evidence workflows for a web app.
SKILL.md
4.0 KB, as published. Nobody here has run it
UI Evidence
Use this skill when the user wants an agent to:
- bootstrap
ui-evidenceinto the current repo - capture
beforeandafterUI screenshots - capture the current UI into a reviewable snapshot bundle
- compare two UI states locally
- generate a review page a human can scan quickly
- persist a reusable config for future UI work
- compare the current checkout against another git ref
This skill is intentionally thin. The CLI is the engine. The skill is the agent-facing entrypoint. The package provides the repo-local executable that actually runs capture and review commands.
Bootstrap first
- Detect the repo package manager from
packageManageror lockfiles. - Check whether the repo can already run
ui-evidence. - If the CLI is missing, install it from GitHub:
pnpm add -D github:0xBrewing/ui-evidence
npm install -D github:0xBrewing/ui-evidence
yarn add -D github:0xBrewing/ui-evidence
bun add -d github:0xBrewing/ui-evidence
- Run the repo bootstrap step with the matching package runner:
pnpm exec ui-evidence install --agent both --config ./ui-evidence/config.yaml
npx ui-evidence install --agent both --config ./ui-evidence/config.yaml
yarn ui-evidence install --agent both --config ./ui-evidence/config.yaml
bunx ui-evidence install --agent both --config ./ui-evidence/config.yaml
That bootstrap should leave repo-local skill copies in .agents/skills/ui-evidence/ for Codex and .claude/skills/ui-evidence/ for Claude Code.
Default flow
- Run
ui-evidence discover. - Read the suggested config and unresolved list.
- Ask only about unresolved values.
- Create or patch
ui-evidence/config.yaml. - Add
ui-evidence/hooks/*only if deterministic state is needed. - Run
ui-evidence doctor, thenui-evidence doctor --readyif route or wait-target confidence is still low. - Run
ui-evidence run,ui-evidence snapshot, orcapture/compare/report/review. - Summarize:
review/index.htmlreport.<lang>.mdmanifest.json- pair and overview images, or current snapshot captures and overview images
Commands to prefer
ui-evidence discover --format json
ui-evidence init --interactive --config ./ui-evidence/config.yaml
ui-evidence doctor --config ./ui-evidence/config.yaml
ui-evidence doctor --config ./ui-evidence/config.yaml --ready --profile mobile-en
ui-evidence run --config ./ui-evidence/config.yaml --stage <stage-id>
ui-evidence run --config ./ui-evidence/config.yaml --stage <stage-id> --before-ref main
ui-evidence run --config ./ui-evidence/config.yaml --stage <stage-id> --after-attach http://127.0.0.1:3000 --resume
ui-evidence snapshot --config ./ui-evidence/config.yaml --scope <scope-id>
ui-evidence review --config ./ui-evidence/config.yaml --stage <stage-id>
ui-evidence review --stage <stage-id> reuses the latest snapshot current captures when a stage has no before/after comparison artifacts yet. If neither stage artifacts nor snapshot artifacts exist, it should fail instead of emitting an empty review.
What to inspect locally first
- package manager and dev scripts
- Playwright config and existing auth states
- harness routes or storybook routes
- stable
data-testidtargets - existing screenshot folders
- whether the repo can boot from another git ref
Rules
- Prefer this skill as the install entrypoint when it is available through
skills add. - Use the package-manager-native runner for every
ui-evidencecommand. - Keep stage definitions stable and additive.
- Prefer existing harness routes over fragile live flows.
- Prefer
testIdwaits over loose selectors. - Treat
review/index.htmlas the default human-facing artifact. - Keep hooks small and deterministic.
- Do not re-implement browser automation in prompt text.