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/rhi-zone/normalize/claude-mdgit clone --depth 1 https://github.com/rhi-zone/normalizeWhat 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 | $0.07124 | $0.07124 |
| Opus 5 | $0.03562 | $0.03562 |
| Sonnet 5 | $0.01425 | $0.01425 |
| Haiku 4.5 | $0.00712 | $0.00712 |
Grade C, and why
normalize CLAUDE.md scanned grade C with 2 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 2d 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.
Downloads and executes remote codehighSupply chain
curl | sh runs whatever the server returns today, which is not necessarily what it returned when this was reviewed.
**installer URL:** `curl -fsSL https://rhi.zone/normalize/install.sh | sh` — the canonical copy lives at `https://github.com/rhi-zone/rhi.zone/blob/master/normalize/install.sh`; the in-repo `install.sh` is just a synced Makes network callslowCapability
Not a fault in itself. Listed so you know the mod talks to something, and to what.
**installer URL:** `curl -fsSL https://rhi.zone/normalize/install.sh | sh` — the canonical copy lives at `https://github.com/rhi-zone/rhi.zone/blob/master/normalize/install.sh`; the in-repo `install.sh` is just a synced How it starts
The opening of the file, as written. The whole thing — 183 lines — stays where its author put it; the contents beside it link to each section on GitHub.
CLAUDE.md
behavioral rules for me in this repo.
stuff to go read when it's relevant: docs/philosophy.md (design tenets), docs/architecture-decisions.md (technical choices), docs/cli-design.md (CLI surface + principles), docs/audit-2026-03-12.md (architecture audit w/ action items).
publishing
it's published on crates.io as 49 crates (+4 publish = false ones: normalize-grammars, normalize-semantic-facts, xtask, benches). all at v0.3.2 rn (early, still actively being built).
installer URL: curl -fsSL https://rhi.zone/normalize/install.sh | sh — the canonical copy lives at https://github.com/rhi-zone/rhi.zone/blob/master/normalize/install.sh; the in-repo install.sh is just a synced copy of it.
API-first
normalize IS an API that happens to have a CLI on top of it. the service layer returns typed data, the CLI just renders it. so when i'm designing a command i start with the data model — what shape does the result actually have? the CLI surface (subcommand name, flags, positional layout) follows from THAT. i never let how it looks in a terminal drive the data shape.
what that means in practice:
- a command that returns a list of items returns
Vec<T>(or a wrapper), no matter whether the input was a flag, a glob, or a subcommand name. --json/--jq/--jsonlare first-class on every command, bc programmatic consumers (agents, scripts, LSP) are primary users, not an afterthought.- when designing a report struct, the question is "what does whoever's calling this API want to DO with the result?" — never "what should this look like printed in a terminal?"
architecture
crate-level context lives in docs/crates.md — that's the canonical registry of every workspace crate (purpose, category, namespace ownership). it replaced the old per-directory SUMMARY.md thing at the crate level. the actually-maintainable source of truth for a crate's purpose is its Cargo.toml description field — keep THAT accurate and the registry stays cheap to regenerate. check the registry before asking "wait which crate owns X?"
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.
- 2d ago First seen · 183 lines · 7,124 tokens per session scan C 4fed3d65acb1
normalize CLAUDE.md is an instructions file published in the GitHub repository rhi-zone/normalize (5 stars, last pushed 4d ago), licensed Apache-2.0. It adds 7,124 tokens to every session, about $0.0356 per session on Opus 5. A static security scan graded it C with 2 findings (downloads and executes remote code, makes network calls). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.
Other instructions, from other repositories
reducethemtokens CLAUDE.md
Instructions for yttrium400/reducethemtokens, covering claude.md, commands, install for development, run tests and build and publish.
Ix CLAUDE.md
Instructions for ix-infrastructure/Ix, covering project: ix (this repo), layout, commands, cli dev (from ix-cli/) and skill + tooling (from repo root).
tree-sitter-fsharp AGENTS.md
Instructions for ionide/tree-sitter-fsharp, covering improving the f# tree-sitter parser, project structure, running the cli, workflow for adding a new feature and 1. create a test case first.
tree-sitter-analyzer CLAUDE.md
Instructions for aimasteracc/tree-sitter-analyzer, covering ruflo — claude code configuration, rules, test quality rules — locked (2026-06-28), t-1: no new test files for existing plugins (blocker) and t-2: no weak assertions — ratchet-enforced (blocker).
tree-sitter-analyzer AGENTS.md
Instructions for aimasteracc/tree-sitter-analyzer, covering agent instructions, codemap index, test runtime contract, ci test tier contract and agent dogfood feedback loop.
kin GEMINI.md
Instructions for firelock-ai/kin, covering kin: read the repository from the graph, pick the tool by what you know, trust the envelope, not the emptiness, if the tools have no graph and a working order.