Borrowing it
Nothing to install: this file belongs to flpccabral/mcp-browser. Take a copy, put it at the same path in your own repository, and replace the rules that are about this project with yours.
curl -O https://raw.githubusercontent.com/flpccabral/mcp-browser/main/.claude/skills/browser-mcp-campanha-confiabilidade-do-agente/SKILL.mdgit clone --depth 1 https://github.com/flpccabral/mcp-browserWrote 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/flpccabral/mcp-browser/browser-mcp-campanha-confiabilidade-do-agente)<a href="https://agentmods.dev/skills/flpccabral/mcp-browser/browser-mcp-campanha-confiabilidade-do-agente"><img src="https://agentmods.dev/badge/skills/flpccabral/mcp-browser/browser-mcp-campanha-confiabilidade-do-agente/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/flpccabral/mcp-browser/browser-mcp-campanha-confiabilidade-do-agente"><img src="https://agentmods.dev/badge/skills/flpccabral/mcp-browser/browser-mcp-campanha-confiabilidade-do-agente.svg" alt="Reviewed on agentmods" width="80" 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 5.1 | $0.00278 | $0.06418 |
| Opus 5 | $0.00139 | $0.03209 |
| Sonnet 5 | $0.00056 | $0.01284 |
| Haiku 4.5 | $0.00028 | $0.00642 |
Grade A, and why
browser-mcp-campanha-confiabilidade-do-agente 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 11d 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 — 377 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Campanha: confiabilidade do agente LLM
Este é o problema mais difícil em aberto do repositório. O dono declarou: a
CONFIABILIDADE do agente (src/browser_mcp/agent.py, classe BrowserAgent, loop
execute_task OBSERVE→THINK→CHECK→ACT→RECORD) é o alvo. O SOTA público
(browser-use, 104k stars) reporta 87.4% em 200 tarefas do benchmark Odysseys
(dado externo, 2026-07-13). Toda esta campanha existe para trocar "acho que
melhorou" por um número.
Linha de base estabelecida em 2026-07-18 (ver baseline-2026-07-18.md):
com provedor rápido e sem rate limit (gemma4:31b-cloud), 7/7 (100%) estável
no conjunto offline, 0 incidentes de parse. A "não-confiabilidade" dos runs
anteriores (Ollama lento, NVIDIA free rate-limited) era confound de provedor,
não defeito do agente. Fase 1 desnecessária; Fase 2 (retry/backoff) implementada
e validada como upside puro; o conjunto de 7 tarefas está saturado — a próxima
alavanca é ampliar o benchmark com tarefas mais difíceis para achar o teto real.
Regra de ouro (leia antes de qualquer coisa)
Nenhuma mudança no agente é promovida sem número de benchmark ANTES e DEPOIS. Sucesso é MENSURÁVEL, nunca a olho. Prompt-tweaking sem benchmark é caminho cercado (ver seção "Caminhos cercados").
Toda promoção passa pelo gate de [[browser-mcp-controle-de-mudancas]] e não pode
introduzir NOVAS regressões. Contagem verificada 2026-07-17:
.venv/bin/python -m pytest tests/ --collect-only -q lista os testes coletados;
algumas suítes de smoke são flaky por timeout de rede (httpbin/networkidle). O
critério é: sem novas regressões além dessa flakiness conhecida de rede, não
"tudo verde cegamente".
Quando NÃO usar esta skill
- Um bug pontual quebrou em runtime (click não funciona, extensão não conecta,
timeout esperando elemento) →
[[browser-mcp-playbook-de-depuracao]](triagem sintoma→causa). Esta skill é campanha sistemática, não triagem. - Você quer entender POR QUE o agente é assim (isTrusted, @e refs, histórico
de design) →
[[browser-mcp-arqueologia-de-falhas]]e[[browser-automacao-referencia]]. - Você vai só rodar o servidor / conectar extensão →
[[browser-mcp-executar-e-operar]]. - Você quer comparar com concorrentes ou escrever claim público →
[[browser-mcp-fronteira-e-posicionamento]]. - A mudança não toca o agente (nova tool, config) → outras skills.
What ships with it
11 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.
- baseline-2026-07-18.md 5.6 KB
- scripts/benchmark_agent.py 17 KB runs code
- scripts/fixtures/accordion.html 1.3 KB
- scripts/fixtures/cascade_ajax.html 4.0 KB
- scripts/fixtures/extract_fact.html 575 B
- scripts/fixtures/form_login.html 1.0 KB
- scripts/fixtures/form_validate.html 1.3 KB
- scripts/fixtures/iframe_fact.html 609 B
- scripts/fixtures/modal_confirm.html 1.2 KB
- scripts/fixtures/pagination.html 1.6 KB
- scripts/fixtures/search_list.html 1.1 KB
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.
- 11d ago First seen · 377 lines · 278 tokens per session scan A 43e0052bb637
browser-mcp-campanha-confiabilidade-do-agente is a skill published in the GitHub repository flpccabral/mcp-browser (0 stars, last pushed 1mo ago), licensed MIT. It adds 278 tokens to every session and 6,418 once invoked, about $0.0014 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 2026-08-31.
Other skills, from other repositories
browse
Drive a real browser through Aside: open a page, read it, click through a flow, take screenshots, check console errors. (gstack).
playwright-cli
A command-line tool for controlling Chromium, Firefox, and WebKit browsers, including navigation, page interaction, screenshots, PDFs, and recorded actions.
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.
create-verification-skill
Create a repo-local verification skill and exhaustive feature map for driving a real app through its UI, CLI, or API. Use for "create a verification skill", "make a verify skill for this repo", or "document how agents can verify this app".
browser-qa
A browser-based quality check for deployed web pages and user flows. It uses browser automation to test rendering, navigation, forms, interactions, responsive behaviour, and accessibility-related issues.
test-electron-app
Drive the real running PostHog Electron app (live tRPC, workspace-server, real data) over CDP with agent-browser. Connect to the running app on port 9222, test desktop changes against a local Django stack, snapshot the accessibility tree, inspect network requests, and screenshot only when explicitly asked. Use when…