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/vercel-labs/dev3000/agents-mdgit clone --depth 1 https://github.com/vercel-labs/dev3000What 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 | $0.01320 | $0.01320 |
| Opus 5 | $0.00660 | $0.00660 |
| Sonnet 5 | $0.00264 | $0.00264 |
| Haiku 4.5 | $0.00132 | $0.00132 |
Grade A, and why
dev3000 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.
How it starts
The opening of the file, as written. The whole thing — 119 lines — stays where its author put it; the contents beside it link to each section on GitHub.
AGENTS.md
This file is the source of truth for agent guidance in this repo. CLAUDE.md is symlinked here.
Runtime
- Use
d3kas the default local runtime. - Canonical Portless HTTPS URLs are required by default and must be genuinely port-free (for example,
https://my-app.localhost). Before starting d3k, rund3k portless status --json; if setup is required, rund3k portless setup, wait for the one-time administrator authorization, and verify Portless again. Do not start the app or browser until canonical Portless is ready unless the user explicitly requests--no-portless. - When the user asks to use, test, debug, or develop with d3k, start it in a retained background tool session. Do not make the user launch it or switch to the TUI.
- Do not run
bun run devorbun run buildforwww/. - Start d3k with:
d3k -t - Primary debugging commands:
d3k status --json d3k errors --context d3k logs -n 200 d3k logs --type browser d3k logs --type server - Drive the monitored browser with:
The session's CDP port is auto-connected. To target a different session, rund3k agent-browser snapshot -i d3k agent-browser click @e2d3k agent-browser connect <port>first.
Local UI, Production Workflows
- Local
www/is for the UI only. - Workflow execution, workflow orchestration, and sandbox activity must run in production, not in the local Next server.
- On localhost, workflow API routes must proxy to production instead of starting workflows or managing sandboxes in-process.
- When testing locally, treat
localhostas a production-backed shell:- start runs from the local UI
- let production own workflow startup, execution, retries, and completion
- let production own sandbox creation, sandbox control, and sandbox logs
- let report pages and runs pages read workflow state from production APIs or shared remote storage
- Do not add new local-only workflow shortcuts, localhost-only workflow execution paths, or tmp-cache fallbacks unless the user explicitly asks for them.
- Do not make localhost the source of truth for workflow state.
- If local workflow storage is ever needed for debugging, gate it behind an explicit env var and keep it off by default.
- If a workflow feature behaves differently locally than in production, prefer removing the local special case rather than extending 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.
- yesterday First seen · 119 lines · 1,320 tokens per session scan A 70f791423514
dev3000 AGENTS.md is an instructions file published in the GitHub repository vercel-labs/dev3000 (1,570 stars, last pushed 5d ago), licensed MIT. It adds 1,320 tokens to every session, about $0.0066 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
terminal-browser AGENTS.md
Instructions for zenbu-labs/terminal-browser, a project described as: A browser inside your terminal.
agent-inspect AGENTS.md
Instructions for rajudandigam/agent-inspect, covering agentinspect ai maintainer instructions, product boundary, source of truth, public-copy rule and start every task.
inspector copilot-instructions.md
Copilot instructions for modelcontextprotocol/inspector, covering copilot review instructions — mcp inspector, typescript, react and ui (web client), the .withprops() rule and state and effects.
jenkins-cli CLAUDE.md
Instructions for avivsinai/jenkins-cli, covering claude.md, build & test commands, run a single test, skip e2e tests during unit testing and e2e with colima on macos (if docker is unreachable).
dile-components AGENTS.md
Instructions for Polydile/dile-components, covering development guidelines for dile components, repository overview, packages, documentation site and demos.
patchloom AGENTS.md
Instructions for patchloom/patchloom, covering agents.md, project overview, dev commands, git hygiene and release prs (release-please).