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/hyperlogue/r3/agents-mdgit clone --depth 1 https://github.com/hyperlogue/r3What 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 | $0.13771 | $0.13771 |
| Opus 5 | $0.06885 | $0.06885 |
| Sonnet 5 | $0.02754 | $0.02754 |
| Haiku 4.5 | $0.01377 | $0.01377 |
Grade A, and why
r3 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 — 805 lines — stays where its author put it; the contents beside it link to each section on GitHub.
r3 — Review. Revise. Resolve.
A local-first review tool for AI-generated code and docs. A long-running per-user daemon on localhost owns review + feedback state for all your repos; reviews are created from the CLI/agent, and you review the commits, diffs, and raw files they capture in the browser and leave line/quote-anchored feedback; an AI agent (or you) replies by id and the decision shows up live over SSE. The daemon, CLI, and SPA ship as one self-contained binary.
For usage see README.md. This file is the orientation map for
working in the repo and the source of truth for its design, together with the
deep-reference skills it points at (see Deep reference) — when a
design decision changes, update the relevant section here or in the skill that owns
it.
Why r3 / prior art. When an AI agent writes code or docs you want to read the result, mark the exact spots you care about, hand those notes back, and watch the agent react in place — without copy-pasting transcripts. The MIT-licensed difit and diffx informed the design; r3's delta from both: a persisted review + feedback/reply model, raw-file (not just diff) reviews, headless CLI creation with queryable session/meta, an agent re-anchor API to keep feedback from orphaning, and a structured reply/watch protocol that round-trips live into the UI.
Architecture
The server is authoritative; there are three clients of one HTTP/JSON API
— the browser (SPA), the CLI, and the agent (through the CLI). Because the CLI and
agent are first-class, the HTTP/JSON contract in shared/types.ts is the
product, not an implementation detail of the React client.
┌── browser (SPA) ─ fetch + SSE ──┐
agent ── CLI (thin HTTP client) ─ HTTP ───────┼──► daemon (Hono + bun:sqlite)
└── you at the terminal ────────────┘ one per user, one port,
one global store
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 · 805 lines · 13,771 tokens per session scan A 569888fb9362
r3 AGENTS.md is an instructions file published in the GitHub repository hyperlogue/r3 (23 stars, last pushed 6d ago), licensed MIT. It adds 13,771 tokens to every session, about $0.0689 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
claude-code-ultimate-guide AGENTS.md
AGENTS.md instructions for FlorianBruniaux/claude-code-ultimate-guide, covering claude code ultimate guide - project context, purpose, repository structure, published at: https://cc.bruniaux.com/whitepapers/ and key files.
claude-code-ultimate-guide CLAUDE.md
Claude Code instructions for FlorianBruniaux/claude-code-ultimate-guide, covering claude code ultimate guide - project context, purpose, repository structure, published at: https://cc.bruniaux.com/whitepapers/ and key files.
wmux CLAUDE.md
Instructions for openwong2kim/wmux, covering language (owner decision, 2026-07-20), skill routing, design system and versioning & release (owner decision, 2026-07-05).
claude-code-ultimate-guide copilot-instructions.md
Copilot instructions for FlorianBruniaux/claude-code-ultimate-guide: Use /CLAUDE.md as the single source of truth for this repository.
dear-hiring-manager CLAUDE.md
Instructions for extrasmall0/dear-hiring-manager, covering claude.md — dear-hiring-manager, what this is, architecture principle, layout and user data (never in repo).
wove copilot-instructions.md
Instructions for mits-pl/wove, covering wave terminal — copilot instructions, project rules, skill guides and preview server.