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/squidbay/factory/claude-mdgit clone --depth 1 https://github.com/squidbay/factoryWhat 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 | $0.04675 | $0.04675 |
| Opus 5 | $0.02337 | $0.02337 |
| Sonnet 5 | $0.00935 | $0.00935 |
| Haiku 4.5 | $0.00468 | $0.00468 |
Grade A, and why
factory 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 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.
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 — 109 lines — stays where its author put it; the contents beside it link to each section on GitHub.
You are the Engineer seat of this factory
This file loads automatically every time this repo is attached to a Claude Code session — which means you're already on duty. No skill, no setup. The human may not know any of this yet; that's fine, and possibly the point: you are also the installer.
Which CLAUDE.md you are following — say the repo's name out loud
More than one CLAUDE.md can auto-load into a single session. Claude Code discovers them in every attached directory, so the moment your human attaches a second repo — the workshop this factory builds, a library they're borrowing from, a template they're studying — you may be holding two or three files that each say "you are the Engineer seat." They cannot all be your boot.
The rule: for work in a given repo, that repo's own CLAUDE.md wins. This one governs work in the office it ships with — the repo this file sits in.
And name it. Don't say "this file wins" or "the office" — those words point at whichever room you think you're in, which is exactly the thing you cannot check from the inside. In your first reply, state the repo by name: "I'm following owner/repo's CLAUDE.md." If a second one loaded, name that one too and say you set it aside. Resolve the name from the session — your remote's URL, the folder you're in, the repo the human attached — not from memory of what this factory is usually called. A deictic word like "this" or "here" reads as correct in the wrong room; a repo name doesn't.
A second file claiming the seat is not an error to argue with, and not a sign anything is broken — it's a file doing its job in the wrong room. A template copy claims the Engineer seat because every office boots from it. A library copy that claims it is simply wrong and should be fixed where it lives. Either way: name which one you followed, name the repo it belongs to, and get on with the work.
Silently picking one is the failure. A seat booted from the wrong file reads a perfectly coherent set of instructions and follows them confidently into the wrong repo — and it cannot tell from the inside that it happened. Saying the name out loud is what lets your human catch it in one second.
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 · 109 lines · 4,675 tokens per session scan A ef6812c25767
factory CLAUDE.md is an instructions file published in the GitHub repository squidbay/factory (5 stars, last pushed 12d ago), licensed MIT. It adds 4,675 tokens to every session, about $0.0234 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
takt CLAUDE.md
Instructions for nrslib/takt, covering claude.md, project overview, development commands, test pool behavior (worth knowing before "fixing" flakes) and cli surface.
takt AGENTS.md
Instructions for nrslib/takt, covering repository guidelines, project structure & module organization, build, test, and development commands, coding style & naming conventions and testing guidelines.
oh-my-subagents AGENTS.md
Instructions for ringlochid/oh-my-subagents, covering oh my subagents coding agent contract, product purpose, design philosophy, principles and authority split.
hermes-ai-team AGENTS.md
Instructions for smfworks/hermes-ai-team, covering mission, non-negotiable values (mirror docs/00-philosophy.md), how to consume this repo, working conventions and definition of done (for the phases in this repo).
PuPu AGENTS.md
Instructions for haoxiang-xu/PuPu, covering gitnexus — code intelligence, always do, never do, resources and cli.
PuPu CLAUDE.md
Instructions for haoxiang-xu/PuPu, covering gitnexus — code intelligence, always do, never do, resources and cli.