agentleFS
Kripu77Kripu77/software-factory/AGENTS.md

software-factory

AGENTS.md7 starsChanged 3 months agoMITView on GitHub

You are one lane in a factory. The runner (Grok, Claude Code, Codex, Cursor) does not change the rules. Read, in order, whatever exists: Use glossary terms from CONTEXT.md. Do not invent synonyms. Missing CONTEXT.md: continue silently. /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).…

  • Reads credentials
  • Where files are headingNew AGENTS.md files have more rules about keeping secrets out of the code, and fewer about code style.

    Of the AGENTS.md files we read, the 31,774 first seen in the last six months, against the older ones: the kind of rule that gained most share, and the kind that lost most. About files like this in general, not this one, and an estimate, since a kind is a pattern in the words.

  • In this repository10 other files it gives its agents.

The file

We split this AGENTS.md into the rules it gives an agent

8 rules. About 0 are common, 5 are distinctive, and 3 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈5

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 3

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  • could be automatic: a setting could enforce it
  1. Distinctive:

    Read, in order, whatever exists:

    line 7
  2. Very specific:

    Use glossary terms from CONTEXT.md.

    line 14
  3. Distinctive:

    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 …

    line 26
  4. Distinctive:

    Commit one complete thought at a time.

    line 28
  5. Very specific:

    Never merge. Never gh pr merge.

    line 33
  6. Very specific:

    Never read .env or .env.local.

    line 34
  7. Distinctive:

    Never attribute factory code to other products.

    line 35
  8. Distinctive:

    Only the ticket or PR you were given.

    line 36

≈ “Common” is an estimate: at least 1 in 5 of the 40,246 AGENTS.md files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
AGENTS.md

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:

  1. CONTEXT-MAP.md at the workspace root
  2. This repo's CONTEXT.md
  3. ADRs
  4. 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 Closes until 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 .env or .env.local. Never store .env contents in factory memory.
  • Never attribute factory code to other products.
  • Only the ticket or PR you were given. No sweeping other people's PRs.
flowchart TD
  after[After-state mermaid] --> page[Existing docs page]
  after --> face{User-facing?}
  face -->|yes| readme[README or QUICKSTART]
  face -->|no| page
  after --> arch{Missing architecture page?}
  arch -->|yes| addPage[Add the page]
  arch -->|no| page
  before[Bug before-state] --> pr[PR only]
  review[Review] --> gap{After-state belongs in docs and is missing?}
  gap -->|yes| changes[Request changes]
  gap -->|no| bulk{Whole-ticket dump?}
  bulk -->|yes| changes
  bulk -->|no| comments[Review comments]

Lanes

  • Tech lead: classify, quiz, ticket, dispatch via factory.sh (usually floor). Lanes report back. Do not implement. Implementing in the lead session is a failed run.
  • Telemetry: evidence only. Report back to Tech lead. Do not implement. Do not classify.
  • Feature: do not expand the ask. Report back to Tech lead when the PR exists.
  • Bug: browser-repro if web, then TDD. Read factory memory at start. Write started if none in progress, then done, blocked, or failed. Report back to Tech lead.
  • Docs: docs only. Report back to Tech lead.
  • Review: comments, not patches. A review summary may use headings. Request changes when required mermaid diagrams are missing, when an after-state belongs in docs and is missing, when a PR body is one wall of text, a line starts with #n, a fence or list has no blank line in front of it, or the PR is one bulk commit: one commit that is the whole ticket. Report the verdict back to Tech lead. Do not merge.
  • CI: checks until green. Report back to Tech lead. Do not merge.
  • QA: smoke or browser-walk. Report findings to Tech lead. Do not implement. Do not merge.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.