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 instructions/hyperyond/hover/claude-mdgit clone --depth 1 https://github.com/Hyperyond/HoverWhat 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 | $0.05049 | $0.05049 |
| Opus 5 | $0.02524 | $0.02524 |
| Sonnet 5 | $0.01010 | $0.01010 |
| Haiku 4.5 | $0.00505 | $0.00505 |
Grade A, and why
Hover CLAUDE.md 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 — 190 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Directory guide
This file is the single source of truth for agents entering the Hover repository. Read this file first. It describes the current implementation and the boundaries agents must respect when working in it.
Core documentation index
- Product scope and onboarding:
README.md. - Architecture and protocols: this file (
CLAUDE.md),packages/core/README.md. - License:
LICENSE(Apache-2.0).
What Hover is
Hover is an MCP server (@hover-dev/mcp) that the user adds to their own coding agent (Claude Code, Cursor, …). The user's agent drives the user's debug Chrome over CDP through Hover's grounded actuation tools, then crystallizes the session into a plain @playwright/test .spec.ts file under __vibe_tests__/. Hover bundles no AI runtime and manages no model — the calling agent is the intelligence (BYO-CLI). Install globally, then register the bin:
npm i -g @hover-dev/mcp
claude mcp add hover -- hover-mcp
The differentiator vs. Stagehand / Midscene / Playwright codegen is record == replay: grounded actuation means the selector that drove the action IS the one saved, and crystallization is deterministic — a string-template translation of the recorded grounded-step buffer, with no LLM in the codegen path. The saved artifact is plain Playwright that runs in CI with no agent in the loop.
Direction (MCP-first, 2026-06)
Hover pivoted from a VS Code extension that was the engine — a chat that spawned a sandboxed agent plus an in-extension staged engine (host.mjs, a WS client pool, a packed copy of core) — to MCP-first: the engine plugs into the user's own agent over MCP. A standalone interactive TUI that was briefly built has been deleted. The VS Code extension was gutted into a passive review cockpit that drives no agent and ships no engine.
Earlier history (for context only): a pre-extension surface was a dev-server-injected widget shipped via bundler plugins (vite-plugin-hover, @hover-dev/astro / nuxt / next, webpack-plugin-hover, @hover-dev/cli, @hover-dev/widget-bootstrap, @hover-dev/transform-source). All of those were removed long ago; their source is gone and only git history retains it.
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 · 190 lines · 5,049 tokens per session scan A 8551f0b43f72
Hover CLAUDE.md is an instructions file published in the GitHub repository Hyperyond/Hover (11 stars, last pushed 1mo ago), licensed Apache-2.0. It adds 5,049 tokens to every session, about $0.0252 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 instructions, from other repositories
browser-control AGENTS.md
Instructions for keon/browser-control: Use this project as a tiny browser pipe, not an agent framework.
rustwright AGENTS.md
Instructions for Skyvern-AI/rustwright, covering repository instructions, public boundary, pull-request review, local verification and browser launches.
chrome-agent CLAUDE.md
Instructions for sderosiaux/chrome-agent, covering chrome-agent v0.15.0, architecture, build & test, release and → requires npmtoken in github secrets.
neobrowser AGENTS.md
Instructions for pitiflautico/neobrowser, covering agents.md — orientation for ai agents & contributors, build, run, test (rust — primary), how it works (rust module map), conventions and ci and release gates.
barebrowse CLAUDE.md
Instructions for hamr0/barebrowse, covering dev rules and project specifics.
codex-browser-bridge AGENTS.md
Instructions for DeliciousBuding/codex-browser-bridge, covering agents.md, 仓库级 skill, mcp 工具设计规范, 1. browser 层（src/browser.rs） and 2. mcp 层（src/mcp/ 目录）.