openpaperwork
AGENTS.md0 starsChanged 11 days agoApache-2.0View on GitHub
Instructions for coding agents working in this repository. Claude Code, Codex, Cursor, Antigravity and other agents read this file; people can read it too. OpenPaperwork is a library of document standards that rides the agent the person already has: they point the agent at a folder, ask in chat, and get editable Word, Excel, PowerPoint, PDF and HTML files back, each checked against the published rules. This repository is that library: the packs, the checks, the plugins and the doors…
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 31,843 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 repository24 other files it gives its agents.
- llms.txtllms.txt
- academic-referencesSkill
- annual-accountsSkill
- business-deckSkill
- business-letterSkill
- _checksSkill
- contract-draftingSkill
- diagramSkill
- docxSkill
- e-invoiceSkill
- employment-statementSkill
- html-deckSkill
- literature-searchSkill
- meeting-minutesSkill
- paperSkill
- pdfSkill
- powerpointSkill
- requirements-specSkill
- _sharedSkill
- social-posterSkill
- trade-documentsSkill
- whatsapp-exportSkill
- xlsxSkill
- zotero-librarySkill
The file
We split this AGENTS.md into the rules it gives an agent
9 rules. About 2 are common, 5 are distinctive, and 2 are very specific to this project.
≈2
Common · estimate
Many files like this have a rule of this kind.
≈5
Distinctive · estimate
This author's own choices. Worth a look.
2
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 15
When someone asks you to set up OpenPaperwork, do these steps and report what passed, what failed and the exact versions used.
- Distinctive: line 18
Check for Python 3.10 or later.
- Distinctive: line 23
Check which agent the person uses: the Claude app or Claude Code, the ChatGPT app or Codex, Cursor, Antigravity.
- Common: line 29
Never ask for an API key in chat, and never write a key into a file in this repository.
- Very specific: line 50
After changing anything in skills, rebuild the packs with python3 scripts/build-portable-pack.py and commit plugins and portable-pack with …
- Distinctive: line 52
Run the checks closest to your change and say which ones you did not run.
- Common: line 72
Use invented or redistributable files.
- Distinctive: line 78
No telemetry, automatic updates or hosted sharing turned on by default.
- Distinctive: line 81
Never add a sign-in for an agent.
≈ “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.
Files
OpenPaperwork
Instructions for coding agents working in this repository. Claude Code, Codex, Cursor, Antigravity and other agents read this file; people can read it too.
OpenPaperwork is a library of document standards that rides the agent the person already has: they point the agent at a folder, ask in chat, and get editable Word, Excel, PowerPoint, PDF and HTML files back, each checked against the published rules. This repository is that library: the packs, the checks, the plugins and the doors that install them, with no account and no agent of its own. A good change makes finished file work better for someone who is not technical.
Set up OpenPaperwork
When someone asks you to set up OpenPaperwork, do these steps and report what passed, what failed and the exact versions used.
- Check for Python 3.10 or later. Install it only after the person agrees.
- From the repository root, link the packs and set their Python up:
sh portable-pack/install.sh, thensh portable-pack/doctor.sh --fix. Run the doctor again until it printsReady. Show the person any line it prints forsudoand let them run it. - Check which agent the person uses: the Claude app or Claude Code, the ChatGPT app or Codex, Cursor, Antigravity. Each signs in on the person's own account, in its own app; never sign in for them. See docs/community-preview/BRING_YOUR_OWN_AGENT.md.
- Optional: prove one file task with
standards(below).
Never ask for an API key in chat, and never write a key into a file in this repository. Stop and ask before anything that costs money.
Where things live
| Area | Purpose |
skills | How the agent makes and revises Word, Excel, PowerPoint, PDF and HTML files, and the pack checks (skills/_checks) |
standards | The library's standards: guidelines, invented cases and scripted checks |
plugins, portable-pack | The packs built for installing: one plugin per department, and the whole library as one; built from skills by scripts/build-portable-pack.py, never edited by hand |
docs/community-preview | Philosophy, agents, standards and contribution guidance |
scripts | The pack builds, starter templates and the merge check |
Checks
- Library:
python3 -m unittest -v standards/test_library.py. - Case protocol:
python3 -m unittest -v test_protocol.pyfromstandards. - The pack checks' end state:
python3 -m unittest -v skills/_checks/tests/test_end_state.py. - Portable pack:
python3 -m unittest discover -s portable-pack/tests -p "test_*.py". - Everything a maintainer runs before a merge:
sh scripts/merge-check.sh .(needsuv).
After changing anything in skills, rebuild the packs with
python3 scripts/build-portable-pack.py and commit plugins and portable-pack with it.
Run the checks closest to your change and say which ones you did not run.
Prove a file task
python3 standards/run.py prepare brief-001 /absolute/path/new-run
# run OpenPaperwork on new-run/task only, save the result under new-run/first/
python3 standards/run.py revision /absolute/path/new-run
python3 standards/run.py check /absolute/path/new-run first
python3 standards/run.py check /absolute/path/new-run revisionThe agent doing the task must not see this repository, the checker or the revision request. A passing check is not acceptance; a person still reviews the rendered file. Details: standards/README.md.
Rules
- One purpose per pull request. Say what the person asked, what they got before and after, and what friction remains. See CONTRIBUTING.md.
- Use invented or redistributable files. No credentials, customer files or private logs in code, fixtures, issues or pull requests.
- Anything copied from another project adds its notice to
THIRD_PARTY.mdandlicenses/in the same change. - Credits for the projects we build on stay in
THIRD_PARTY.mdanddocs/community-preview/CREDITS.md. - No telemetry, automatic updates or hosted sharing turned on by default.
- This tree has no agent of its own; do not add one. The library rides the agent the person already has.
- Never add a sign-in for an agent. Each agent signs in on its own, on the person's own account.
- Account sign-in, hosted services, team coordination, connections and built-in registrations stay out of this tree.
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
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.