DOCFORGE

INSERT REPOSITORY. GENERATE DOCUMENTATION. NO INVENTED LORE.

► PRESS START

---

An Agent Skill cartridge for AI coding agents: designs, writes, audits, and maintains documentation grounded in the actual source — code graphs, manifests, git history, repository evidence. Every claim carries a blob-stamped provenance, every document passes an independent audit before it counts.

PRECHECK → ANALYZE → PLAN → WRITE → AUDIT → TRACK

Docforge reads the repository's own size and shape before deciding how big the docs should be: a small repo gets compact layout — the same subjects, folded into fewer, denser files — while a large one gets the full standard tree; either way you can override the suggestion. Documents that fall out of scope on a later run (tier downgrade, dropped profile, layout switch) are retired, not deleted out from under you — moved to a git-ignored .docforge/obsolete/<year>/ (or removed, if you say so explicitly) with the history kept in the manifest.

██▓▒░ INSERT COIN ░▒▓██

# Agent Skills — insert in the current project
npx skills add jonaskahn/docforge

# or globally
npx skills add jonaskahn/docforge -g -y

Claude Code native marketplace:

/plugin marketplace add jonaskahn/docforge
/plugin install docforge@docforge

Both paths load skills/docforge — the required core bundle. docforge-revise and docforge-dashboard are thin optional entrypoints on top of it.

██▓▒░ HOW TO PLAY ░▒▓██

Invocation order is always command → scope args → flags. Plain language works across compatible agents:

Document this repository from scratch.
Create diligence documentation for this service.
Generate ADRs from the repository history.
Check which generated docs have drifted from source.

/docforge — fresh start

Intake, plan, or write a tree from scratch. On intake it performs only read-only discovery, then asks its scope questions in two turns: first your goal and the documentation layout, then tier, repository profiles, audience, graph source, and execution mode. It finally summarizes the complete scope and asks you to confirm, edit, or cancel before writing anything.

Command	Use
/docforge	Fresh start: intake, plan, or write
/docforge --plan-only	Plan / dry-run tree only

/docforge-revise — keep it current

Re-ground what drifted; a bare run only syncs manifest metadata.

Command	Use
/docforge-revise	Migrate/upgrade manifest metadata only — no questions, no writing
/docforge-revise all	Full-tree structural refresh
/docforge-revise <area>	Scoped revise (architecture, flows, operations, …)
/docforge-revise flow	Flow harvest → organize → derive → write
/docforge-revise flow --plan-only	Flow analysis only (no body writes)

/docforge-dashboard — liquid docs

Render the written docs as a local Fumadocs site (Liquid Glass theme).

Command	Use
/docforge-dashboard	dashboard start: reconcile → rebuild when changed → serve → open
/docforge-dashboard --plan-only	Preflight, metadata dry-run, signatures, route plan
/docforge-dashboard export	Static export (next build → <dashboard>/out/) for GitHub Pages / S3
/docforge-dashboard status	Read-only: signature match, server state, document count
/docforge-dashboard stop	Stop the detached dev server

CHEAT CODES

--plan-only — analyze / dry-run tree only. --auto-accept — skip routine pauses after the scope confirm. --no-dashboard — skip the automatic dashboard build/serve at completion. --help — per-command reference (canonical text in help.md).

██▓▒░ DEMO ░▒▓██

The cabinet's attract mode after a run:

██▓▒░ SYSTEM REQUIREMENTS ░▒▓██

• a compatible AI coding agent
• git
• one tool runtime: Python 3.10+ or Node.js 22+ / Bun / Deno
• Node.js 22+ and npm only for the dashboard install/serve steps

Full procedure and policy: workflows, document contracts, provenance model.

██▓▒░ CREDITS ░▒▓██

Created by Jonas Kahn. SPDX-License-Identifier: MIT — LICENSE. Bugs: issues. Contributions: pull requests.

GAME OVER? NO. CONTINUE. →