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/dipakkr/agentsync/agents-mdgit clone --depth 1 https://github.com/dipakkr/agentsyncWhat 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 | $0.00706 | $0.00706 |
| Opus 5 | $0.00353 | $0.00353 |
| Sonnet 5 | $0.00141 | $0.00141 |
| Haiku 4.5 | $0.00071 | $0.00071 |
Grade A, and why
agentsync 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 — 53 lines — stays where its author put it; the contents beside it link to each section on GitHub.
AgentSync — instructions for coding agents working in this repo
You are one member of a team of humans and AI agents collaborating on this repo through a shared AgentSync hub (chat + task board + presence). Follow this on every session so you never collide with a teammate.
1. First thing: identify yourself
If .agentsync/identity.json does NOT exist, you have not been introduced yet.
Ask the human, then register:
"Who am I working as? I need: your name, this machine's label, and which agent I am (claude / codex / kimi / …)."
Then call the MCP tool agentsync_register with those values. Your member id is
<person>.<machine>.<agent> (e.g. deepak.mac-a.claude). After that, the hub knows
you and you appear on the dashboard roster.
If .agentsync/identity.json already exists, read it and reuse that identity — do
not ask again.
2. Before you write any code
get_plan— there must be an approved plan. If none, help the human draft one (tasks, each with a file scope) rather than coding blind.list_tasks— pick an unclaimed task that matches your role.claim_task— this locks the task's file scope to you. If the hub warns of an overlap with another member's active claim, STOP and coordinate in chat first.
3. While you work
- Work only inside your claimed task's file scope + your own git worktree/branch
(
<person>/<agent>/<slug>). post_messagewith akind:fyi(default status — nobody replies),ask(a question; setto, expects one reply),reply(answer an ask; setreply_to). Postfyiwhen you start, finish, or get blocked. Leavetoempty to tell everyone.read_messagesto receive — the hub never pushes chat at you, you pull it. It returns your inbox plusneeds_reply(asks addressed to you that nobody answered). Reply only to those; never reply to fyi, to a reply, to your own messages, or tosystemnotices, and don't send acks — that's what prevents ping-pong loops. After anask,wait_for_messageblocks for the answer. Pass the lastmax_idassince_id. Check it when you start, when blocked, before push.- Commit small and often. Never commit protected paths (
.env,*.key,secrets.*).
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 · 53 lines · 706 tokens per session scan A 836268c5d967
agentsync AGENTS.md is an instructions file published in the GitHub repository dipakkr/agentsync (0 stars, last pushed 24d ago), licensed MIT. It adds 706 tokens to every session, about $0.0035 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
martin-loop AGENTS.md
Instructions for Keesan12/martin-loop, covering martinloop agent operating rules, repository authority, mandatory internal-first sequence, prohibited actions and required internal health proof.
Kiln AGENTS.md
AGENTS.md instructions for Kiln-AI/Kiln, covering project overview, project goals, tech stack, agent tools and agent prompts.
liveblocks AGENTS.md
AGENTS.md instructions for liveblocks/liveblocks, covering project structure, bash commands, code style, dependencies and code quality.
open-tag AGENTS.md
Instructions for fancyboi999/open-tag, covering open-tag — guide for ai coding agents, this file is a map, not a manual, conventions, parallel development (worktrees) and isolated dev e2e (on demand).
sample-well-architected-skills-and-steering copilot-instructions.md
Copilot instructions for aws-samples/sample-well-architected-skills-and-steering, covering well-architected framework, role, when to apply, pillars and design principles.
cli AGENTS.md
AGENTS.md instructions for jumbocontext/cli, covering agents.md and instructions for agents on how to collaborate with jumbo.