About ops-browser
ops-browser is a skill for carrying out browser operations within defined task authority. It emphasizes verifying the target, separating observation from side effects, and reporting evidence, outcomes, and gaps.
Key features
- 01Verify browser, profile, tab, and target
- 02Keep read-only observation distinct from writes
- 03Capture and compare visual evidence
- 04Scope DevTools and source evidence checks
Use cases
- 01Inspecting a specified browser page
- 02Verifying a local workspace target
- 03Comparing visual states
- 04Debugging with DevTools or extracting design evidence
Install
Copy the command and run it in a terminal.
npx skills add https:/After installing, type/ops-browserin Claude Code to confirm it loaded.
Bundled files
References and scripts installed along with SKILL.md.
agents/openai.yamlassets/frontend-visual-evidence.schema.jsonreferences/axure-product-evidence.mdbrowser-operation-protocol.mddevtools-debugging.mdeval-cases.mdfrontend-visual-evidence.mdlanhu-ui-evidence.mdlocal-browser-workspaces.mdplatform-operations.mdtab-lifecycle.mdusage.mdviewport-policy.mdscripts/preflight-local-browser-workspace.pyresolve-local-browser-route.pyvalidate-frontend-visual-evidence.pyDetails
From the project READMEBrowser Operations
Entry Gate
Operate the requested browser surface within existing task authority. A request to run and inspect a local result authorizes its ordinary read-only acceptance; do not ask again merely because implementation hands off here. Preflight browser/profile/tab, target URL/state and applicable account/side-effect boundary. Stop only the operation whose target or authority is unresolved. External sends and writes retain their gates.
Route Map
| Request condition | Read | Result |
|---|---|---|
| Any browser operation | platform operations, tab lifecycle, and viewport policy | Bounded target operation at a verified viewport |
| External-AI turn, attachment/submission, or structured operation handoff | Applicable sections of browser operation protocol | Correlated side-effect and response evidence |
| Local workspace/browser ownership applies | local workspaces | Verified target selection |
| Runtime visual comparison applies | visual evidence | Same-state capture/comparison |
| DevTools/console/network debugging applies | DevTools debugging | Scoped observation |
| Axure/Lanhu evidence extraction applies | Axure evidence and/or Lanhu evidence | Qualified source evidence |
Invariants
- Inventory and identify the exact target before action; absent target with no creation policy is not verified.
- Separate read observation from side effects; never infer identity, authorization, or application acceptance from browser reachability.
- Preserve target/profile and avoid focus theft; report actual postcondition and unverified gaps.
Output Map
Return surface/profile/tab identity, target state, authorized action, evidence/captures, postcondition, cleanup state, and gaps.
Reference Map
- Read platform operations, tab lifecycle, and viewport policy for target and side-effect invariants. Load the browser operation protocol only for external-AI turns, attachments/submissions, or structured handoffs; ordinary page inspection needs no provider-turn ledger.
- Read local workspaces, visual evidence, DevTools debugging, Axure evidence, and Lanhu evidence only when applicable.
- Read usage when selecting a backend or mode, performing a write, capture, recording, or download, or producing a protocol handoff. Maintainers only: read eval cases; do not load it during ordinary runtime.