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 skills add open-gsd/gsd-browser --skill gsd-browser-skillgit clone --depth 1 https://github.com/open-gsd/gsd-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/open-gsd/gsd-browser/gsd-browser-skill)<a href="https://agentmods.dev/skills/open-gsd/gsd-browser/gsd-browser-skill"><img src="https://agentmods.dev/badge/skills/open-gsd/gsd-browser/gsd-browser-skill.svg" alt="Measured on agentmods" height="20"></a>- NVIDIA SkillSpector pass
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 | $0.00175 | $0.02032 |
| Opus 5 | $0.00088 | $0.01016 |
| Sonnet 5 | $0.00035 | $0.00406 |
| Haiku 4.5 | $0.00017 | $0.00203 |
Grade A, and why
gsd-browser 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 8d 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 — 133 lines — stays where its author put it; the contents beside it link to each section on GitHub.
<essential_principles>
Primary path for AI agents (2026+): Use gsd-browser mcp (the stdio MCP server). It exposes 50+ tools, live resources (latest-snapshot, current-state, active-recordings, timeline, etc.), and executable prompts (robust_login_flow, full_page_audit, autonomous_research_task, evidence_creation_workflow, debug_stuck_agent_flow, etc.) with rich agent-optimized envelopes (summary + structured_data + suggested_next_actions + evidence_refs).
For MCP clients (Cursor, Claude Desktop, VS Code + Copilot, etc.): After connecting to gsd-browser mcp, read the root docs/AGENT-BEST-PRACTICES.md (golden rules, workflows, "When to Use What" table) and docs/mcp.md. Run ./scripts/mcp-quickstart.sh cursor (or claude/vscode) for tailored config.
The daemon auto-starts on browser commands. gsd-browser daemon health reports state and does not start a session. Use gsd-browser daemon start only when you want to pre-warm or verify daemon lifecycle explicitly.
Always re-snapshot after page changes. Refs are versioned (@v1:e1, @v2:e3). After navigation, form submission, or dynamic content loading, old refs are stale. Run gsd-browser snapshot (or read the gsd-browser://latest-snapshot MCP resource) to get fresh refs before interacting with _ref tools.
Use --json when parsing output. Use text mode when reading output yourself. Use --json when you need to extract values programmatically. MCP tools return rich structured envelopes.
Positional args have no flag prefix. Commands like click, type, hover take positional args. Do NOT add --selector:
gsd-browser click "button.submit"(correct)gsd-browser click --selector "button.submit"(WRONG)
Core workflow pattern: Every browser automation follows: navigate -> snapshot (or read latest-snapshot resource) -> interact (prefer semantic act/find_best first, then refs) -> re-snapshot (after DOM changes). For complex flows, prefer browser_batch via MCP.
gsd-browser navigate https://example.com
gsd-browser snapshot
# Read snapshot output: @v1:e1 [input type="email"], @v1:e2 [button] "Submit"
gsd-browser fill-ref @v1:e1 "[email protected]"
gsd-browser click-ref @v1:e2
gsd-browser wait-for --condition network_idle
gsd-browser snapshot # REQUIRED - old refs are now stale
Command chaining: Use && when you don't need intermediate output. Run separately when you need to parse output first (e.g., snapshot to discover refs, then interact). MCP agents: use browser_batch for atomic multi-step execution.
Use the live viewer when the user wants to watch or direct the browser. gsd-browser view opens a localhost viewer with live frames, narrated action history, ref overlays, and pause/step/resume/abort controls + annotation + record capabilities. Keep using CLI/MCP commands for actions; the viewer is the shared screen and control surface. MCP tools: browser_view, browser_takeover, browser_annotation_request, browser_record_start/stop, browser_goal, browser_step, etc. These are among the highest-leverage collaborative features.
Global options available on all commands:
| Flag | Purpose |
|---|---|
--json |
Structured JSON output |
--browser-path <path> |
Path to Chrome/Chromium |
--cdp-url <url> |
Attach to an already-running Chrome instance |
--session <name> |
Named session for parallel instances + persistent action cache / state (highly recommended for MCP agents) |
--no-narration-delay |
Skip narration lead-time sleeps while keeping history/events |
</essential_principles>
Based on what the user needs, read the appropriate workflow:
| User intent | Workflow |
|---|---|
| Navigate, click, type, fill forms, interact with pages (MCP or CLI) | workflows/navigate-and-interact.md |
| Share the browser screen, narrate actions, pause/step/resume/abort, human collaboration + evidence | workflows/live-viewer-and-narration.md |
| Scrape data, extract content, read page structure | workflows/scrape-and-extract.md |
| Test pages, run assertions, visual regression, mock network | workflows/test-and-assert.md |
| Debug issues, check logs, diagnose problems (use debug_bundle + MCP prompts) | workflows/debug-and-diagnose.md |
| Install, configure, set up sessions, MCP client setup | workflows/setup-and-configure.md |
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.
- references/command-reference.md 13 KB
- references/configuration.md 2.6 KB
- references/error-recovery.md 3.0 KB
- references/semantic-intents.md 2.8 KB
- references/snapshot-and-refs.md 3.4 KB
- workflows/debug-and-diagnose.md 3.4 KB
- workflows/live-viewer-and-narration.md 4.2 KB
- workflows/navigate-and-interact.md 4.7 KB
- workflows/scrape-and-extract.md 4.7 KB
- workflows/setup-and-configure.md 5.7 KB
- workflows/test-and-assert.md 6.2 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.
- 8d ago First seen · 133 lines · 175 tokens per session scan A f3d8b354d5b3
gsd-browser is a skill published in the GitHub repository open-gsd/gsd-browser (47 stars, last pushed 3d ago), licensed Apache-2.0. It adds 175 tokens to every session and 2,032 once invoked, about $0.0009 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
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.
debug-optimize-lcp
Guides debugging and optimizing Largest Contentful Paint (LCP) using Chrome DevTools MCP tools. Use this skill whenever the user asks about LCP performance, slow page loads, Core Web Vitals optimization, or wants to understand why their page's main content takes too long to appear. Also use when the user mentions…
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.
google-safe-browsing
Prevent and fix Google Safe Browsing "Dangerous site" flags. Use when launching a public web app, buying/picking a domain, building a login or signup page, or when any site shows a red "Dangerous site" / "Deceptive site" warning in Chrome, Brave, Safari, Firefox, or Edge. Triggers on "dangerous site", "deceptive…