Docs Forge: AI-Powered Automated Documentation Generation Workflow

Documentation maintenance is a long-standing pain point in software development—code evolves continuously, but documentation often lags behind. The Docs Forge project attempts to fundamentally solve this problem by building a cross-platform agent workflow. Its core design principles include code-first, knowledge accumulation, minimal intervention, and traceable sources. It supports multiple AI programming assistants like Codex and Claude Code, uses a seven-stage process to generate structured engineering documents, adapts to various mainstream frameworks, and significantly improves the accuracy and maintainability of documentation.

The core pain point of documentation maintenance lies in the contradiction between rapid code iteration and delayed documentation updates. Docs Forge is positioned as a documentation generation skill for multi-platform AI programming assistants, rather than a simple document template tool. Its design philosophy transforms the documentation generation process from arbitrary creation to fact-based organization, effectively solving the problems of documentation accuracy and maintainability.

Docs Forge has unique cross-platform adaptation capabilities:

• Codex: Available as a plugin market installation
• Claude Code: Available as a native SKILL.md skill
• Antigravity: Via AGENTS.md and optional GEMINI.md adapters
• Other programming assistants that support reading AGENTS.md The multi-platform strategy covers the mainstream AI programming assistant ecosystem, allowing users to get a consistent experience without switching tools.

Docs Forge uses a structured seven-stage process with clear inputs and outputs for each stage:

1. Scope Confirmation: Determine document type, target audience, framework selection, output path, and existing document handling strategy
2. Code Ingestion: Inventory the codebase and classify non-skipped files (full read, sample read, statistics only, explicitly skipped)
3. Gap Analysis: Identify questions that code cannot answer (e.g., reasons for design decisions, unreflected architectural constraints)
4. Intent Consultation: Only ask maintainers the questions identified in the gap analysis
5. Document Generation: Generate framework-native documents based on the knowledge base and supplementary answers, which must include source code references
6. Visual Capture: Run the application to capture screenshots or videos after explicit consent
7. Assembly and Delivery: Build navigation, validate links, and generate an integrity report

Docs Forge adapts to various mainstream documentation frameworks: Fumadocs, Docusaurus, Mintlify, Nextra, Starlight, MkDocs Material. If the target repository has no framework configured, the agent will detect the tech stack and recommend appropriate options. Before generating documents, a structured knowledge base is established, including project overview, architecture documents, public interfaces, functional features, existing document analysis, build and run instructions, glossary, and pending confirmation questions, ensuring the process is transparent and auditable.

Applicable Scenarios:

• Architecture documents, API references, SDK/library documents
• Configuration and environment variable references, deployment guides
• Contributor onboarding guides, operation and maintenance manuals, troubleshooting
• Product user guides, tutorials, and walkthroughs

Non-Applicable Scenarios:

• Individual docstrings or code comments
• Short README paragraph edits
• Blog posts unrelated to the codebase

Design Constraints: Allow reading/writing document files, modifying adapters and outputs; prohibit fabricating public APIs, exporting public interfaces without confirmation, silently overwriting existing documents, and running external services without consent—ensuring AI assistance does not introduce false information.