Doc Yet

Doc Yet

Mission

Turn a real repository into durable, navigable documentation.

• Keep the canonical internal documentation in .agents.
• Improve the root README.md in place when it exists.
• Generate cross-tool bridge files so Cursor, Claude, Codex, and Gemini can all find the same source of truth.
• Interpret the repository shape first. Do not force the same folder layout onto every project.

Core workflow

1. Inspect the repository before proposing a documentation structure.
2. Detect the project topology: single app, feature-based app, monorepo, multi-service repo, shared package workspace, or mixed architecture.
3. Ask a few short questions only where choices are still unresolved.
4. Run a gap analysis against existing docs before writing anything.
5. Create or update the minimum useful documentation set in .agents.
6. Improve the root README.md without flattening all details into it.
7. Generate or refresh bridge files for Cursor, Claude, Codex, and Gemini.
8. Summarize what was created, what was updated, and what remains uncertain.

Questions to ask after the first inspection

Ask short, concrete questions. Prefer 3-6 questions. Use defaults when the user does not answer all of them.

Always ask:

• Which AI tools are actively used in this repo? Default to Cursor, Claude, Codex, and Gemini.
• Who is the main audience for the docs right now: new contributors, current team, AI agents, or external integrators?
• Are there any folders, services, or features that should be excluded from documentation?

Ask when relevant:

• For monorepos: which apps or packages matter most right now?
• For frontend repos: is there a design system, token source, or Storybook-like source of truth that should be documented?
• For backend repos: are there external APIs, queues, workers, or scheduled jobs that must be covered?
• For security-sensitive code: should secrets, internal URLs, or operational details be intentionally omitted?

If the user is unsure, continue with reasonable assumptions and record them in the docs.

Repository inspection checklist

Inspect before writing:

• Root manifests and workspace manifests: package.json, pnpm-workspace.yaml, turbo.json, nx.json, Cargo.toml, go.mod, pyproject.toml, etc.
• App and package boundaries: apps/, packages/, services/, src/features/, modules/, libs/.
• Entrypoints, build tooling, framework config, and deployment config.
• Existing docs: root README.md, docs/, existing .agents/, ADRs, architecture notes.
• API signals: route definitions, OpenAPI files, GraphQL schemas, RPC contracts, client SDKs.
• Data signals: ORM models, schema files, migrations, seed scripts, warehouse models.
• Infra signals: Docker, compose, CI, Terraform, Helm, Vercel, Render, Fly, AWS, GCP, Azure.
• Frontend signals: component libraries, Storybook, theme files, CSS variables, Tailwind config, token JSON, design-system packages.
• Auth and security signals: auth providers, RBAC code, middleware, policy files, permission maps.

Do not document guessed architecture as fact. Mark assumptions clearly.

Canonical documentation layout

Use .agents as the canonical home for internal project documentation, but adapt placement to repository shape.

Global docs

For the whole repository, keep the canonical global docs in:

• .agents/project/documentation-index.md
• .agents/project/project-overview.md
• .agents/project/architecture.md
• .agents/project/tech-stack.md
• .agents/project/gap-analysis.md

Local scoped docs

When a subtree has enough independence or complexity, create a local .agents inside that subtree as well.

Common examples:

• apps/web/.agents/
• apps/api/.agents/
• packages/ui/.agents/
• services/billing/.agents/
• src/features/analytics/.agents/
• src/domains/orders/.agents/

Use local .agents when the subtree has one or more of these properties:

• Independent ownership or release cadence
• Distinct runtime or deployment model
• Non-trivial public surface area
• Enough internal complexity that future agents would benefit from local context
• A shared UI or design-system boundary

Do not create a local .agents for every folder. Prefer fewer, well-linked docs over documentation sprawl.

Topology rules

Read references/topology-playbook.md before deciding placement.

High-level defaults:

• Single-repo, single-deployable app: keep most docs in .agents/project/.
• Monorepo: always create .agents/project/, then add local .agents for meaningful apps, packages, or services.
• Feature-based frontend: create .agents/project/, then local .agents for major features or domains when they are large enough to deserve their own map.
• Shared UI/design-system package: add a local .agents there and document component conventions and tokens.
• Multi-service backend: global docs plus local .agents for each important service.

When unsure, prefer this progression:

1. Global .agents/project/
2. Local .agents only for high-value boundaries
3. Cross-links between global and local indexes

Required outputs

Always create or update:

• .agents/project/documentation-index.md
• .agents/project/project-overview.md
• .agents/project/architecture.md
• .agents/project/tech-stack.md
• .agents/project/gap-analysis.md
• Root README.md

If README.md exists, improve it in place. If it does not exist, create it in English by default.

Use Mermaid diagrams in architecture.md by default when they materially help explain flows, boundaries, or integrations.

Optional outputs

Read references/documentation-blueprints.md and generate the subset that matches the repo:

• repository-map.md
• api-contracts.md
• data-model.md
• integrations.md
• auth-and-security.md
• deployment-and-environments.md
• testing-strategy.md
• observability.md
• business-rules.md
• onboarding.md
• troubleshooting.md
• glossary.md
• adr/
• design-system.md
• design-tokens.md

Frontend-specific expectation

If the repo contains a frontend with shared UI primitives, themes, tokens, or component patterns, strongly prefer documenting:

• design-system.md
• design-tokens.md
• Relevant local feature docs when the app is feature-based

README language rule

• Preserve the dominant language of the existing README.md.
• If the current README is mostly Portuguese, keep it in Portuguese when improving it.
• If there is no README, default to English.
• The .agents docs should be written in English unless the user explicitly asks otherwise.

Gap analysis rules

Before writing final docs, create or refresh .agents/project/gap-analysis.md with:

• Existing docs worth preserving
• Missing docs
• Outdated docs
• Proposed local .agents placements
• Open questions
• Assumptions used for the current pass

This file is not a brainstorm dump. Keep it crisp and actionable.

Content rules

• Prefer verified repository facts over generic best-practice filler.
• Use concrete file paths, commands, frameworks, environments, services, and package names.
• Preserve good existing documentation instead of rewriting for style alone.
• Do not create placeholder sections with TODO unless the unknown is important and genuinely unresolved.
• Distinguish clearly between verified facts, likely inferences, and open questions.
• Keep docs navigable: short sections, cross-links, and clear file names.
• Use tables when they improve scanability, not by default.
• When the project is large, summarize details in the global docs and push specifics into local .agents.

README rules

The root README.md is the public entrypoint, not the entire internal wiki.

It should usually contain:

• What the project is
• Main apps, services, or packages
• Quick start
• Key commands
• High-level architecture snapshot
• Where to find deeper docs in .agents

It should usually not contain:

• Every package detail in a monorepo
• Exhaustive internal architecture prose
• Long operational runbooks better kept in .agents

Cross-tool bridge files

By default, generate or update these bridge files:

• AGENTS.md
• CLAUDE.md
• GEMINI.md
• .cursor/rules/doc-yet-project-docs.mdc
• .claude/commands/doc-yet.md

Read references/bridge-templates.md before writing them.

Bridge file rules:

• Keep them thin. They should point to canonical docs instead of duplicating them.
• Point to .agents/project/documentation-index.md first.
• Tell the tool to also read the nearest local .agents/ when working inside a subtree that has one.
• Preserve any strong existing project instructions; merge instead of overwriting blindly.
• Keep update instructions short and operational: after major architecture or workflow changes, refresh .agents and the root README.md.

Do not create additional tool-specific command files unless the repo already uses them or the user asks for them.

Blueprint references

Use these bundled references instead of inventing structure from scratch:

• references/documentation-blueprints.md
• references/topology-playbook.md
• references/bridge-templates.md

Delivery checklist

• The global docs exist in .agents/project/.
• Local .agents directories exist only where they add real value.
• architecture.md includes Mermaid when useful.
• The root README.md was improved or created in the correct language.
• Bridge files were created or carefully updated for Cursor, Claude, Codex, and Gemini.
• Frontend repos with real UI systems include design-system and token docs.
• Monorepos link global and local documentation clearly.
• The docs state assumptions instead of presenting guesses as facts.

Source: Dirosaki/skills — distributed by TomeVault.