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/kripu77/software-factory/agents-mdgit clone --depth 1 https://github.com/Kripu77/software-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 | $0.01052 | $0.01052 |
| Opus 5 | $0.00526 | $0.00526 |
| Sonnet 5 | $0.00210 | $0.00210 |
| Haiku 4.5 | $0.00105 | $0.00105 |
Grade A, and why
software-factory 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 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 — 65 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Software factory
You are one lane in a factory. The runner (Grok, Claude Code, Codex, Cursor) does not change the rules.
Before you explore
Read, in order, whatever exists:
CONTEXT-MAP.mdat the workspace root- This repo's
CONTEXT.md - ADRs
docs/agents/if present
Use glossary terms from CONTEXT.md. Do not invent synonyms. Missing CONTEXT.md: continue silently.
Skills
/tdd, /implement, /unslop for product code. /mermaid when writing a feature, bug, or docs PR.
/code-review only when reviewing.
/loop-on-ci only when watching PR checks.
/to-tickets only when breaking work into GitHub issues.
/qa only when testing a running app (smoke or browser). /run-smoke-tests and /browser-use are the QA skills.
Shipping
- Before writing code, find a similar, recently-merged example in this repo (a file or PR doing the same kind of thing) and match its patterns. If none exists, say so in the PR description.
- New branch off
main. Do not wreck other local branches or worktrees. - Commit one complete thought at a time. The message names that thought. Keep the diff small. If the thought would be huge, split the ticket. No line-count cap.
- PR title:
Type/<issue.number>/<short description>where Type is Feat, Bug, Arch, Chore, Refactor, or General. - PR description: human, at most 3 sentences, then mermaid. Feature: before and after when a prior shape exists, after-only when net-new. Bug: before and after. Use short paragraphs with a blank line between them. A blank line before any list or fence. Cite the ticket as a markdown link. A line never starts with #n. Do not
Closesuntil every slice has landed. - After-state mermaid also lands in the product repo docs tree in the same PR. Update the existing page for that subsystem. Do not add a new file per PR. If the bug revealed a missing architecture page, add that page and persist only the corrected after-state. README or QUICKSTART only when the diagram is user-facing: first-run, signal path, or a flow a human runs. Internals stay in docs. Bug before-state stays on the PR.
- New issue comments a lane writes use the same spacing. Ledger mem-write comments stay one sentence, a blank line, then a markdown PR link. Do not rewrite old PRs.
- Never merge. Never
gh pr merge. Never--no-verify. - Never read
.envor.env.local. Never store.envcontents in factory memory. - Never attribute factory code to other products.
- Only the ticket or PR you were given. No sweeping other people's PRs.
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 · 65 lines · 1,052 tokens per session scan A b306f4c4abcf
software-factory AGENTS.md is an instructions file published in the GitHub repository Kripu77/software-factory (7 stars, last pushed 3d ago), licensed MIT. It adds 1,052 tokens to every session, about $0.0053 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
claude-code-karma CLAUDE.md
Instructions for JayantDevkar/claude-code-karma, covering claude.md, project overview, quick start, clone (no submodule init needed — this is a monorepo) and start api (terminal 1).
rn-dev-agent AGENTS.md
Instructions for Lykhoyda/rn-dev-agent, covering repository guide for agents, repository map, editing rules, architecture rules and supported node runtimes.
cybrix-skills CLAUDE.md
Instructions for cybrixcc/cybrix-skills, covering cybrix-skills, what this repo is, install command (for reference), structure and key rules.
yet-another-agentic-chat CLAUDE.md
Instructions for amyodov/yet-another-agentic-chat, covering claude.md, vocabulary, layout, verified facts and message format.
qector-claude-plugin AGENTS.md
AGENTS.md instructions for GuillaumeLessard/qector-claude-plugin, covering policylayer playbook, repository guidelines, project structure & module organization, build, test, and development commands and coding style & naming conventions.
vscode buildNext.instructions.md
Working notes and architecture documentation for the new esbuild-based build system in build/next. Use when making changes to the new build pipeline (transpile/bundle commands, NLS plugin, source-map handling, resource copying, or self-hosting watch tasks).