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/monkilabs/opencastle/chrome-devtoolsnpx skills add monkilabs/opencastle --skill chrome-devtoolsgit clone --depth 1 https://github.com/monkilabs/opencastleWhat 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 | $0.00074 | $0.00568 |
| Opus 5 | $0.00037 | $0.00284 |
| Sonnet 5 | $0.00015 | $0.00114 |
| Haiku 4.5 | $0.00007 | $0.00057 |
Grade A, and why
browser-testing 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 2d 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 — 34 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Browser Testing with Chrome DevTools MCP
Project test app, selectors, suites, and breakpoints: testing-config.md. Docs: https://developer.chrome.com/docs/devtools
Context budget — the main constraint
Screenshots are expensive. MAX 3 per session, reserved for failures. Assert with evaluate_script instead — element counts, window.location.href, !!document.querySelector(...), textContent, new URL(location.href).searchParams.toString(). take_snapshot (DOM) is far lighter than take_screenshot. One focus area per session; clear browser state between unrelated flows.
Tools
navigate_page—{ type: 'url', url }or{ type: 'reload' }click/type/wait_for—clickandtypetake auidfrom a prior snapshot, not a CSS selectorevaluate_script—{ function: '() => ...' }(an arrow function string, not a raw expression)resize_page—{ width, height }list_console_messagesperformance_start_trace—{ reload: true, autoStop: true }; thenperformance_analyze_insight({ insightSetId, insightName })
wait_for timing out almost always means the dev server is down or the URL is wrong — check that before debugging selectors.
Workflow
Navigate → wait_for anchor text → assert via evaluate_script → exercise interactions → hit an edge case URL (e.g. ?q=nonexistent) and assert the empty state → list_console_messages (any error: fix source, rebuild, reload, restart from navigate) → re-run at every breakpoint, verifying interactions and not just layout. Most layout bugs only appear at narrow viewports.
Regression re-test
Read the prior result.json, build + lint, then re-run the entire previous suite — a fix routinely regresses a different test. Every test must pass before writing the updated result.json. Do not stop on partial green.
What ships with it
1 file 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.
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.
- 2d ago First seen · 34 lines · 74 tokens per session scan A af022d9d87c4
browser-testing is a skill published in the GitHub repository monkilabs/opencastle (61 stars, last pushed 4d ago), licensed MIT. It adds 74 tokens to every session and 568 once invoked, about $0.0004 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
product-architect
Complete product development system with 64 agents and 35 frameworks. Use when the user wants to build a product, write a PRD, plan an MVP or roadmap, design an app, research a market or check whether a feature already exists or is novel, do competitive analysis, run a security audit, build a financial model, plan…
codex-delegation
Use when a coding task would benefit from delegating work to Codex in the background — a deep independent second opinion, security or architecture analysis, or a parallel implementation running while the session continues. Lets Claude drive the codex companion itself (task, review, status --wait, result) without the…
cursor-delegation
Use when a coding task would benefit from delegating work to Cursor in the background — fast parallel implementation, scaffolding, or an everyday review running while the session continues. Lets Claude drive the cursor companion itself (task, review, status --wait, result) without the user typing /cursor: commands.
cursor-cli-runtime
Internal helper contract for calling the cursor-companion runtime from Claude Code.
cursor-prompting
Internal guidance for composing cursor-agent prompts and picking models for coding, review, diagnosis, and research tasks inside the Cursor Claude Code plugin.
cursor-result-handling
Internal guidance for presenting Cursor helper output back to the user.