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/naticha/bunwv/claude-mdgit clone --depth 1 https://github.com/NatiCha/bunwvWrote 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/instructions/naticha/bunwv/claude-md)<a href="https://agentmods.dev/instructions/naticha/bunwv/claude-md"><img src="https://agentmods.dev/badge/instructions/naticha/bunwv/claude-md.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 | $0.02125 | $0.02125 |
| Opus 5 | $0.01063 | $0.01063 |
| Sonnet 5 | $0.00425 | $0.00425 |
| Haiku 4.5 | $0.00213 | $0.00213 |
Grade B, and why
bunwv CLAUDE.md scanned grade B with 1 finding 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.
Asks for rootmediumPrivilege escalation
A mod that escalates privileges can change anything on the machine, not only the project.
- Socket + PID files `chmod 0600` on daemon start — other local users can't connect to this session How it starts
The opening of the file, as written. The whole thing — 113 lines — stays where its author put it; the contents beside it link to each section on GitHub.
CLAUDE.md
This is bunwv — a headless browser automation CLI built on Bun.WebView (Bun v1.3.12+). Cross-platform: WebKit on macOS (default), Chrome on Linux/Windows (auto-detected).
Project Structure
bunwv.ts # CLI entry point (shebang, arg parsing, commands)
lib/daemon.ts # WebView daemon (Bun.serve on Unix socket, all routes)
skills/bunwv/ # Claude Code skill file (SKILL.md)
.claude-plugin/ # Marketplace plugin config
docs/ # Bun.WebView API reference notes
Development
bun bunwv.ts help # run locally during dev
bun bunwv.ts start # test the daemon
bun bunwv.ts close # stop it
Default to Bun for everything — no Node, no npm scripts, no external dependencies.
Architecture
- CLI (
bunwv.ts) parses args into flags/positional, sends HTTP requests to the daemon via Unix socket - Daemon (
lib/daemon.ts) owns aBun.WebViewinstance, exposes all WebView methods as HTTP routes on/tmp/bunwv-<session>.sock - Sessions are isolated by name — each gets its own socket, PID file, and WebView instance
- Auto-shutdown after 30min idle (configurable via
--idle-timeout)
Design principles (agent-first)
bunwv is designed for AI coding agents driving it via discrete tool calls. Humans are a secondary user. When the two conflict, agent UX wins.
- Statelessness over streams. Every command is one request, one response, exit. No long-lived connections, no background processes for the agent to manage. "Tail" patterns use snapshot + cursor (
--since <seq>), never streaming. - Token efficiency in the happy path. Successful verbs print only what's needed to continue.
click,type,navigate,press,scroll,resize,back/forward/reload: empty stdout, exit 0. Don't spend tokens confirming success. - Structure only where it earns its tokens. JSON is ~30–50% heavier than prose for small outputs. Use plain text for simple success; use JSON for (a) errors, (b) genuinely structured data (events, console buffer, multi-field status under
--json), (c) anywhere the agent must distinguish types —evaluateprints JSON-literal values so strings keep their quotes and numbers don't. - Reliability on the unhappy path. Errors are always JSON on stderr with nonzero exit. Exit codes are stable and documented. Agents branch on exit code, not on stderr parsing. (0 ok, 1 generic, 2 usage, 3 timeout, 4 element-not-found, 5 daemon-unreachable.)
- Error-level console auto-surfaces; everything else is pull. If
console.errororconsole.warnfires during a verb, the daemon attachesX-Console-Errorsto the HTTP response and the CLI forwards it to process stderr as{console:[…]}. Normal logs, CDP events, and navigation events stay in ring buffers, fetched on demand viabunwv console --since/bunwv events --since. - File paths for binary outputs. Screenshots default to a file path on disk; the agent's file-reading tools handle the image.
--encoding base64and--out -(stdout) exist but are not the default. --jsonfor opt-in uniformity. Agents that prefer one shape over terseness can force an envelope{ok, data?, error?, exitCode}on any command.- 1:1 with Bun.WebView. Every capability of the
Bun.WebViewtype is reachable from the CLI. No JS-API-only escape hatches.
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 · 113 lines · 2,125 tokens per session scan B 5be5a8ad1d6f
bunwv CLAUDE.md is an instructions file published in the GitHub repository NatiCha/bunwv (213 stars, last pushed 3mo ago), licensed MIT. It adds 2,125 tokens to every session, about $0.0106 per session on Opus 5. A static security scan graded it B with 1 finding (asks for root). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other instructions, from other repositories
Browser4 CLAUDE.md
Claude Code instructions for platonai/Browser4, covering browser4 — project context for claude, architecture, key dispatch chain (cli → browser), batch commands and e2e test structure.
cc-websearch CLAUDE.md
Claude Code instructions for Djarvur/cc-websearch, covering project, constraints, technology stack, what not to add and conventions.
fix-quera AGENTS.md
Instructions for AlirezaKeshavarz83/fix-quera, covering agent guidance, project shape, visual style, quera page data findings and compatibility findings.
sniff AGENTS.md
AGENTS.md instructions for Aboudjem/sniff, covering agents.md: sniff, what this repo is, how an agent should use sniff, handling the playwright setup gate and finding output schema.
fast-browser CLAUDE.md
Claude Code instructions for m4ttstack/fast-browser, covering fast browser plugin, where a change belongs, fork branch: use fast-browser-runtime, releasing a new runtime and re-pinning this repo: use the script.
sendblue-browser-use AGENTS.md
Instructions for sendblue-api/sendblue-browser-use, covering agents.md — sendblue-browser-use, what this repo is, setup, common commands and health (no auth).