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/iclawapp/iclaw/agents-mdgit clone --depth 1 https://github.com/iClawApp/iClawWrote 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/iclawapp/iclaw/agents-md)<a href="https://agentmods.dev/instructions/iclawapp/iclaw/agents-md"><img src="https://agentmods.dev/badge/instructions/iclawapp/iclaw/agents-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 | $0.04501 | $0.04501 |
| Opus 5 | $0.02250 | $0.02250 |
| Sonnet 5 | $0.00900 | $0.00900 |
| Haiku 4.5 | $0.00450 | $0.00450 |
Grade A, and why
iClaw 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 4d 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 — 217 lines — stays where its author put it; the contents beside it link to each section on GitHub.
AGENTS.md — Guardrails for AI coding agents working on iClaw
Hi. If you're an AI coding agent (Claude Code, Codex, Cursor, anyone) about to make changes here, read this once before you touch the source.
What iClaw is
A local web UI for OpenClaw Gateway. Nothing more. The app:
- Renders a ChatGPT-style chat surface (sidebar of chats, conversation in the middle).
- Stores chat history locally in SQLite (
data/iclaw.db). - Talks to a running OpenClaw Gateway on the same machine via the native WebSocket protocol (no HTTP chat completions, no SSE).
- Auto-loads the bearer token from
~/.openclaw/openclaw.json. - Adds higher-level features that the gateway doesn't have natively: per-project shared facts, scheduled messages, slash autocomplete, header model picker, today's cost chip, exec approval cards.
What iClaw is not
These are intentional non-goals. Don't add them. If a request lands here that needs any of them, push back or open an issue first.
- Not a generic OpenAI-compat client. We migrated off
/v1/chat/completionsin commita92c10e. Do not re-introduce an HTTP chat completion path or SSE-based streaming — everything chat-related is native WS. - Not a multi-provider chat tool. No fallback to Ollama, vLLM, Claude API, OpenAI direct, etc. OpenClaw is the only backend; switch model providers in
openclaw.json. - Not a remote/hosted product. Loopback-only. No multi-user, no auth layer, no team features. Threat model is "this user on this machine".
- For the default Full Power chat, not a re-implementation of OpenClaw's agent runtime. On that path we don't compact transcripts, run tools, or wrap the LLM — we display what the gateway produces and (separately) maintain a project-scoped fact layer additional to OpenClaw memory. (The Work / Safe work / Incognito modes are the deliberate exception — a small, self-contained agent loop in
packages/iclaw-runtime; see below. Don't grow it into a general OpenClaw replacement.) - Not heavyweight on the frontend. No build step beyond
tsc. Plain CSS, vanilla JS on the client, EJS for views. If you want to add React/Vue/Svelte, open an issue first — the answer is probably no for the current scope.
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.
- 4d ago First seen · 217 lines · 4,501 tokens per session scan A 31897ef8d4a2
iClaw AGENTS.md is an instructions file published in the GitHub repository iClawApp/iClaw (10 stars, last pushed 3d ago), licensed MIT. It adds 4,501 tokens to every session, about $0.0225 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 2026-08-31.
Other instructions, from other repositories
clawlet AGENTS.md
Instructions for mosaxiv/clawlet, covering repository guidelines, project structure, build, test, and development commands, build a local binary and run from source.
manifest CLAUDE.md
Claude Code instructions for mnfst/manifest, covering manifest development guidelines, what manifest is, important: cloud mode always, 1. ensure the postgresdb container is running and 2. create a pristine database with a unique name.
nullclaw AGENTS.md
Instructions for nullclaw/nullclaw, covering agents.md — nullclaw agent engineering protocol, 1) project snapshot (read first), 3) engineering principles (normative), 3.1 kiss and 3.2 yagni.
nullclaw CLAUDE.md
Instructions for nullclaw/nullclaw, covering claude.md, mandatory reference, build & test commands, requires exactly zig 0.16.0 (verify: zig version) and build flags.
openclaw AGENTS.md
AGENTS.md instructions for openclaw/openclaw, covering agents.md, start, repair doctrine, product doctrine and clawsweeper review policy.
openclaw CLAUDE.md
Claude Code instructions for openclaw/openclaw, a project described as: The AI that really does things. Any OS. Any Platform. The lobster way. 🦞.