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/agentconnect-md/agentconnect/claude-mdgit clone --depth 1 https://github.com/agentconnect-md/agentconnectWhat 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 | $0.04189 | $0.04189 |
| Opus 5 | $0.02094 | $0.02094 |
| Sonnet 5 | $0.00838 | $0.00838 |
| Haiku 4.5 | $0.00419 | $0.00419 |
Grade A, and why
agentconnect 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 3d 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 — 268 lines — stays where its author put it; the contents beside it link to each section on GitHub.
CLAUDE.md
This file provides guidance to Claude Code (claude.ai/code) when working with code in this repository.
Product behavior and user-facing invariants are documented in
docs/product-conventions.md. Read it before changing
how AgentConnect presents or delivers messages to users.
What this is
AgentConnect is a multi-agent platform that bridges IM platforms (Slack /
Telegram / Discord) to AI coding agents (Claude, Codex) over ACP. The
authoritative design lives in docs/designs/ — start with
architecture.md, and use the grouped index at
docs/README.md to find the rest. Design documentation and every
visual asset embedded in public documentation are maintained in English. Before
adding or reusing a diagram or screenshot, inspect its visible text and verify
that it still matches the current architecture. Prefer diffable SVG or Mermaid
source over opaque raster diagrams.
The defining architectural choice: the Control Plane is never on the message hot path. Agent execution always happens inside a daemon on the data plane. Where that daemon runs is a deployment choice, not part of the invariant — self-hosted on machines the organization operates, or a member of the install's managed Kubernetes pool (Cloud), which shares one PostgreSQL data plane and launches runtimes in sandbox pods; see §3.1 of the architecture design. Platform ingress is either daemon-owned directly or forwarded by the optional relay when a stable public callback endpoint is required; the CP only orchestrates. Concretely:
- A daemon owns direct platform connections, local routing, provider API egress, and agent execution over ACP the Control Plane never sees (local IPC self-hosted; one in-cluster dial to the sandbox pod in the pool). It also accepts pre-addressed public ingress from the relay. It is a self-contained "message + agent execution unit" and keeps running established sessions even if the CP is down (graceful degradation).
- The optional relay terminates Slack HTTP callbacks, GitHub and GitLab webhooks, generic webhooks, and webchat, then forwards content directly to the owning daemon. It does not persist message content.
- The Control Plane does orchestration/registry/auth + serves the Web UI BFF. It
stores only control-plane metadata — never message bodies, ACP
session/updatestreams, or attachment bytes. Authorized BFF reads may proxy bounded transcript, tool-body, memory, or workspace content from the owning daemon without persisting it. - daemon ↔ CP is a single WebSocket used primarily for control signaling (register, heartbeat, orchestration commands, telemetry). It also carries the scoped request/reply frames for those on-demand BFF reads; live platform messages and ACP update streams never use 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.
- 3d ago First seen · 268 lines · 4,189 tokens per session scan A 2a8e3f80f2fc
agentconnect CLAUDE.md is an instructions file published in the GitHub repository agentconnect-md/agentconnect (868 stars, last pushed 3d ago), licensed Apache-2.0. It adds 4,189 tokens to every session, about $0.0209 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
acp-forge AGENTS.md
Instructions for ky2renzzz/acp-forge, covering acp forge — agent instructions, project overview, architecture, conventions and build & test.
obsidian-agent-client AGENTS.md
AGENTS.md instructions for RAIT-09/obsidian-agent-client, covering agent client plugin - llm developer guide, architecture, data flow, acp event flow (single path) and permission flow.
intelligent-terminal rust.instructions.md
Concise Rust coding conventions for this repository.
sprawling AGENTS.md
Instructions for 2youg1/sprawling, covering agents.md — how work is done in this repository, the loop, read before you write, one change, five steps and the rules a machine holds.
sous-chef AGENTS.md
Instructions for tomascupr/sous-chef, covering sous-chef, map and working agreements.
agiwo AGENTS.md
Instructions for xhwSkhizein/agiwo, covering agents.md, how to use this file, truth source order (adr 0048 / 0049), logging and config hot reload.