Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx agentmods add skills/browser-use/windows-harness/windows_harnessnpx skills add browser-use/windows-harness --skill windows_harnessgit clone --depth 1 https://github.com/browser-use/windows-harnessWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/browser-use/windows-harness/windows_harness)<a href="https://agentmods.dev/skills/browser-use/windows-harness/windows_harness"><img src="https://agentmods.dev/badge/skills/browser-use/windows-harness/windows_harness.svg" alt="Measured on agentmods" height="20"></a>What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5 | $0.00066 | $0.04784 |
| Opus 5 | $0.00033 | $0.02392 |
| Sonnet 5 | $0.00013 | $0.00957 |
| Haiku 4.5 | $0.00007 | $0.00478 |
Grade A, and why
windows-harness scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 5d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 387 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Windows Harness
Use one CLI call per decision point, not per primitive:
windows-harness <<'PY'
app = "Notepad"
win.see(app)
win.type("hello", app=app)
print(win.see(app))
PY
The CLI preloads win, lazy browser, Path, and subprocess. Prefer bounded
stdin programs; reserve windows-harness repl for manual exploration and always
exit it.
How to invoke: exactly two forms
One-off questions — use a subcommand (identical in PowerShell, cmd, and bash):
windows-harness apps # app inventory, compact JSON (--all adds system windows)
windows-harness see "CC Switch" # one screenshot -> JSON with its "path"
windows-harness state "Notepad" # UIA tree as JSON (--screenshot adds image)
Anything multi-step — write a .py file, then run it. No quoting or encoding
traps in any shell; this is always the most reliable channel:
windows-harness run task.py # win, browser, Path, subprocess preloaded
Write generated task scripts to the harness scripts dir (printed by
windows-harness doctor as scripts_dir, default
%USERPROFILE%\.windows-harness\scripts); run resolves bare filenames
there. Never write task scripts into the caller's working directory or
invent ad-hoc folders like .windows-harness-temp/ — they pollute the
user's repo and show up as untracked files. Never run them via
python script.py / python -c either: bypassing windows-harness run
also bypasses the preloaded win session, so every call pays a fresh
interpreter start and loses the element table behind element_index.
Never use <<'PY' heredocs outside bash; never enumerate windows through
hand-rolled C#/PowerShell — win.list_apps() already did it. Screenshots
land in %TEMP%\windows-harness-*.png and persist: the returned "path"
stays readable after the CLI exits, so just open it. Set
WINDOWS_HARNESS_OVERLAY=off to skip the pointer renderer for minimum
latency.
Delivery: foreground first
Every input primitive takes delivery="foreground" (default) or
delivery="background".
What ships with it
14 files beside SKILL.md in the same directory: the scripts, references and assets a skill reads on demand. Not counted in the per-session cost; read them before you install if any of them is executable.
- __init__.py 406 B runs code
- agents/openai.yaml 215 B
- annotation.py 7.0 KB runs code
- browser.py 1.7 KB runs code
- capture.py 30 KB runs code
- cli.py 10 KB runs code
- controls.py 25 KB runs code
- delivery.py 14 KB runs code
- inject.py 70 KB runs code
- overlay_helper.py 7.0 KB runs code
- overlay.py 3.0 KB runs code
- pointer.py 669 B runs code
- telemetry.py 5.7 KB runs code
- windows.py 55 KB runs code
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 5d ago First seen · 387 lines · 66 tokens per session scan A 1d2f0c9e78d9
windows-harness is a skill published in the GitHub repository browser-use/windows-harness (61 stars, last pushed 10d ago), licensed MIT. It adds 66 tokens to every session and 4,784 once invoked, about $0.0003 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other skills, from other repositories
use-agent-browser-for-airi
Test AIRI display-model imports with agent-browser across stage-tamagotchi Electron, stage-web, and stage-pocket mobile web layouts. Use when uploading and verifying contributor-supplied Live2D ZIP, VRM, or MMD ZIP/PMX/PMD files through AIRI's model selector, including onboarding bypass, format-specific import…
opencli-sitemap-author
Use when creating or maintaining OpenCLI site sitemaps: agent-facing navigation, page-state, action, workflow, API-reference, pitfall, and fallback knowledge for a website. Use after browser exploration discovers durable site context, when a sitemap is stale, or when promoting local site knowledge into the repo.
webapp-testing
Toolkit for interacting with and testing local web applications using Playwright. Supports verifying frontend functionality, debugging UI behavior, capturing browser screenshots, and viewing browser logs.
interactive-login
How to complete browser/interactive logins (aws / gh / glab / gcloud). The platform backgrounds the login poller so it survives the human's browser round-trip — and when that does NOT work.
pinchtab-mcp
Use this skill when a task requires browser automation through PinchTab's MCP server connected to a remote browser instance. Covers navigation, element interaction, data extraction, form filling, multi-step flows, and session management via MCP tools.
azure-messaging-webpubsub-java
Build real-time web applications with Azure Web PubSub SDK for Java. Use when implementing WebSocket-based messaging, live updates, chat applications, or server-to-client push notifications.