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/browser-use/desktop/agents-mdgit clone --depth 1 https://github.com/browser-use/desktopWhat 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 | $0.00453 | $0.00453 |
| Opus 5 | $0.00227 | $0.00227 |
| Sonnet 5 | $0.00091 | $0.00091 |
| Haiku 4.5 | $0.00045 | $0.00045 |
Grade A, and why
desktop AGENTS.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 yesterday.
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.
What it actually says
Browser Use Desktop Agent Notes
Local Dev Profiles
- The app's runtime database lives under Electron
userData, not under the git worktree. Do not assume the default profile when working across branches. - Use
task worktree:profile:pathto get this branch's isolated profile path. It resolves to.task/user-data/<current-branch>by default. - Use
task worktree:upto start the app against that isolated profile. - Use
task worktree:profile:clean FORCE=1to delete this branch's isolated profile after quitting the app. - Use
task db:worktree:copy FROM=defaultto copysessions.dbfrom the platform default profile into this branch profile. PassFROM=branch:<name>to copy from another branch profile, including branch names with/. Use an absolute path,./relative/path,~/path, orFROM=path:<path>for explicit filesystem paths. PassFORCE=1if the target DB already exists. - Use
task db:worktree:doctorafter copying or when local task runs fail. It compares the targetsessions.dbschema version and schema ID against the current checkout'ssrc/main/sessions/schema-manifest.json. - If the app was launched with
AGB_USER_DATA_DIR, runtask agent:runwith the sameAGB_USER_DATA_DIR. The local task runner reads<userData>/local-task-server.json; pointing the CLI at a different profile will miss the running app even ifsessions.dbis valid. - Quit the app before copying or cleaning profile files. SQLite WAL files and Electron runtime files can be open while the app is running.
Session Schema Changes
DB_SCHEMA_VERSIONgates migrations at runtime.src/main/sessions/schema-manifest.jsontracks the expected fresh schema ID for CI/main drift detection.- Run
task db:schema:checkafter touchingSessionDbmigrations. - Run
task db:schema:updateonly after an intentional schema change.
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.
- yesterday First seen · 34 lines · 453 tokens per session scan A d02684c15502
desktop AGENTS.md is an instructions file published in the GitHub repository browser-use/desktop (668 stars, last pushed 28d ago), licensed MIT. It adds 453 tokens to every session, about $0.0023 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-harness AGENTS.md
Instructions for browser-use/browser-harness, covering code priorities and contributing.
browser-harness CLAUDE.md
Instructions for browser-use/browser-harness, a project described as: Browser Harness | Self-healing harness that enables LLMs to complete any task.
SmolVM AGENTS.md
Instructions for CelestoAI/SmolVM, covering smolvm context, 🚀 project overview, 🧪 development, key commands and release checklist.
SmolVM CLAUDE.md
Instructions for CelestoAI/SmolVM, a project described as: Open-source AI sandbox infrastructure with unified API for VMMs -- Firecracker, QEMU and libkrun.
browser-use CLAUDE.md
Instructions for browser-use/browser-use, covering claude.md, high-level architecture, core components, event-driven browser management and cdp integration.
ClawBench AGENTS.md
Instructions for TIGER-AI-Lab/ClawBench, covering clawbench -- agent context, what this is, project structure, setup and 2. configure at least one model.