browser-proof
Skill2 starsChanged 2 months agoMITView on GitHub
Prove a user-facing claim in the actual browser with exact candidate context, the real user action, screenshots, console and network checks, and cleanup. Use when the thing being claimed is visible behavior, not just backend output.
In this repository19 other files it gives its agents.
- access-checkSkill
- babysit-prsSkill
- bugfixSkill
- bug-reproSkill
- cleanupSkill
- contextSkill
- dev-flowSkill
- docs-managerSkill
- hot-reloadSkill
- jira-managerSkill
- polishSkill
- pr-managerSkill
- release-proofSkill
- status-reportSkill
- sync-projectSkill
- watch-loopSkill
- whatsapp-meSkill
- workflowSkill
- worktreeSkill
The file
We split this skill into the rules it gives an agent
9 rules. About 0 are common, 9 are distinctive, and 0 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈9
Distinctive · estimate
This author's own choices. Worth a look.
0
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 8
Use the browser when the claim is about what a user sees or does.
- Distinctive: line 13
Use the real entry path and the real user action.
- Distinctive: line 15
Check console and relevant network traffic around the target action.
- Distinctive: line 16
Never substitute an API-only proof for a UI claim.
- Distinctive: line 35
Use the consumer that matters:
- Distinctive: line 41
Do not replace a click with a database update or curl request when the claim is about the user journey.
- Distinctive: line 43
Test the failure boundary that is most likely to hide the defect: wrong role, empty state, validation failure, stale session, duplicate …
- Distinctive: line 49
Do not describe a contract injection as full end-to-end automation.
- Distinctive: line 51
If one layer is unavailable, name the missing proof and lower the verdict.
≈ “Common” is an estimate: at least 1 in 5 of the 350,495 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
Browser Proof
Use the browser when the claim is about what a user sees or does.
Core Rules
- Name the exact candidate before starting.
- Use the real entry path and the real user action.
- Save screenshot evidence from the same run that proves the result.
- Check console and relevant network traffic around the target action.
- Never substitute an API-only proof for a UI claim.
- Clean up temporary browser state created only for the proof.
Workflow
- Record the candidate identity: SHA, build, revision, preview, slot, or URL.
- State the exact claim being proved.
- Open the real page or flow.
- Clear old console and network noise if the tool supports it.
- Perform the real action.
- Capture:
- visible result
- screenshot path
- relevant console state
- relevant network result
- Report
pass,fail, orblocked.
Prove The Real Boundary
Use the consumer that matters:
- UI feature: real browser journey.
- CLI feature: actual CLI command and normal authentication path.
- API-only contract: real client or exact request shape, clearly labeled as API proof.
Do not replace a click with a database update or curl request when the claim is about the user journey. Lower-level checks can explain the result, but they do not become UI proof.
Test the failure boundary that is most likely to hide the defect: wrong role, empty state, validation failure, stale session, duplicate action, page refresh, responsive viewport, or multi-instance delivery when relevant.
Evidence Rules
- Tie the screenshot, console, and network evidence to the same run and time window.
- Record whether the event was produced through the real upstream flow or injected to test a downstream contract.
- Do not describe a contract injection as full end-to-end automation.
- Redact credentials and customer data while keeping candidate, time, status, and resource identity useful.
- If one layer is unavailable, name the missing proof and lower the verdict.
Cleanup
Close the browser session created for the proof. Remove only task-owned temporary auth, test data, downloads, recordings, and port forwards. If cleanup fails, return blocked and list the residual state even when the visible assertion passed.
Output Contract
BROWSER PROOF | claim=<summary> | verdict=<pass|fail|blocked>
CANDIDATE | <sha|build|revision|slot|url>
ACTION | <real user action>
RESULT | <visible outcome>
SCREENSHOT | <path-or-artifact>
CONSOLE | <clean|warning|error summary>
NETWORK | <relevant request summary>
CLEANUP | <done|not-needed|blocked>Stop Conditions
Stop with:
passwhen the real action produced the expected visible result and cleanup is complete;failwhen the journey contradicted the claim;blockedwhen candidate identity, access, browser operation, required evidence, or cleanup cannot be proved.
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.

