スキル一覧に戻る
faxenoff

ultracode-autodoc

by faxenoff

Ultra-fast MCP server enabling AI agents to comprehensively work with TypeScript/JavaScript/Python projects: instant structural search, code modification with automatic linting/formatting/fixes, runtime debugging and backwards error analysis, integrated Git automation. Your AI's survival kit for massive codebases!

4🍴 0📅 2026年1月22日
GitHubで見るManusで実行

SKILL.md


name: UltraCode AutoDoc version: 1.0.0 description: Working with .autodoc/ directory, automatic documentation generation, semantic doc search triggers:

  • autodoc
  • documentation
  • .autodoc
  • doc generation

UltraCode AutoDoc — Claude Code Skill

Auto-activated for: Working with .autodoc/ directory, documentation generation

AutoDoc — automatic code documentation system with semantic search.

Key Features

🔗 Auto-updated Code References

All code references in AutoDoc are automatically kept up-to-date. When code moves or changes, line number references are updated automatically — they're always accurate.

🧠 Project Memory

AutoDoc describes the meaning of your application — processes, scenarios, business logic. Search finds relevant code even when there are no keywords or comments in the code itself. The documentation links directly to code sections.

📝 Human + Auto Hybrid

  • Empty AUTODOC.md files in module folders are auto-generated from code structure
  • You can enrich them with descriptions, scenarios, business context
  • Everything you add is preserved and becomes searchable "project memory"
  • System only updates code line references, never overwrites your content

autodoc_search searches code AND documentation simultaneously. Results are enriched with context from both sources — find code by business meaning, not just keywords.


Concept

AutoDoc ≠ code comments

CommentsAutoDoc
HOW code worksWHY, WHO uses, WHICH scenarios
Local contextBusiness context + architecture
For developer in IDEFor AI + system understanding
Keywords required to findFind by meaning, no keywords needed

.autodoc/ Structure

.autodoc/
├── ARCHITECTURE.md    # System components (create manually)
├── FLOW.md            # Business scenarios (create manually)
├── PROCESSES.md       # Technical processes (create manually)
├── DEPENDENCIES.md    # Packages, APIs, microservices (create manually)
├── DEPLOYMENT.md      # Build, CI/CD, ENV (create manually)
├── GLOSSARY.md        # Terms (create manually)
│
└── src/               # Mirrors code structure
    ├── _index.md      # Directory overview
    ├── module/
    │   ├── AUTODOC.md # Auto-generated, enrichable
    │   └── ...
    └── ...

Note: Top-level documents (ARCHITECTURE.md, FLOW.md, etc.) should be created manually. Module-level AUTODOC.md files are auto-generated but can be enriched. All code references are auto-updated.


Using AutoDoc

If Project Has .autodoc/

ALWAYS use AutoDoc for understanding the project:

  1. Start with autodoc_search — instant semantic search across code + docs
  2. Read .autodoc/ files — business context that helps find code by meaning
FileRead when...
ARCHITECTURE.mdNeed to understand project structure
FLOW.mdNeed to understand business scenarios
PROCESSES.mdNeed to understand technical processes
DEPENDENCIES.mdNeed to understand external dependencies
DEPLOYMENT.mdNeed to understand build, CI/CD
src/**/AUTODOC.mdNeed to understand specific module

If Project Doesn't Have .autodoc/

Suggest to user: "Would you like to enable AutoDoc for this project?" → autodoc_init({ language: 'ru' }) or 'en'

Then:

  1. Auto-generate module AUTODOC.md files: autodoc_generate preview=false
  2. Create top-level docs manually: ARCHITECTURE.md, FLOW.md, etc.
  3. Enrich module docs with business context as you learn the codebase

MCP Tools

Initialization & Status

autodoc_init

Initialize AutoDoc. Sets documentation language.

ParamTypeDefaultDescription
enabledbooleantrueEnable AutoDoc
languageenum"en"en / ru / zh
docsDirstring".autodoc"Documentation directory

autodoc_status

Status: enabled/disabled, statistics (docs, sections, refs).

autodoc_detect_language

Auto-detect language from code comments and existing docs.

ParamTypeDefaultDescription
scopeenum"all"comments / docs / all
sampleSizenumber20Files to sample

Search & Read

autodoc_search

Semantic search across code + documentation. Use instead of regular search — provides context!

ParamTypeDefaultDescription
querystringrequiredSearch query
modeenum"hybrid"text / semantic / hybrid
limitnumber10Max results

autodoc_get

Get documentation by path. If section specified — only that section.

ParamTypeDefaultDescription
filePathstring-File path
docIdstring-Document ID

Save

autodoc_save

Save document. Automatically:

  • Parses references
  • Generates embeddings for semantic search
  • Extracts sections
ParamTypeDefaultDescription
filePathstringrequiredFile path
contentstringrequiredMarkdown content
typeenum-Document type

Validation & Sync

autodoc_validate

Check all references, optionally fix broken ones.

ParamTypeDefaultDescription
fixBrokenbooleanfalseFix broken refs

autodoc_sync

Sync documentation with code changes.

ParamTypeDefaultDescription
scopeenum"outdated"all / outdated / file
directionenum"both"both / disk-to-db / db-to-disk

Auto-generate

autodoc_generate

Auto-generate documentation for the codebase.

ParamTypeDefaultDescription
previewbooleantruePreview without writing
useLlmbooleanfalseUse LLM for descriptions
incrementalbooleantrueOnly update changed
languageenum"auto"auto / en / ru / zh
modulestring-Generate for single module

History

autodoc_changelog

View documentation change history.

ParamTypeDefaultDescription
limitnumber20Max entries
sincetimestamp-Filter since

Workflows

1. Initialize Project Documentation

User: "Document this project"

1. autodoc_detect_language({ scope: 'comments' })
   → Detect language from comments

2. autodoc_init({ enabled: true, language: 'ru' })
   → Initialize AutoDoc

3. autodoc_status()
   → Show current state

4. Create .autodoc/ with documentation
   autodoc_save({ filePath: 'ARCHITECTURE.md', content: '...' })

2. Understand Existing Project

User: "How does authentication work?"

1. autodoc_search({ query: "authentication auth", mode: 'semantic' })
   → Finds documentation + code

2. autodoc_get({ filePath: 'FLOW.md', section: 'registration' })
   → Business scenario

3. autodoc_get({ filePath: 'PROCESSES.md', section: 'auth-flow' })
   → Technical implementation

3. After Code Changes

User: "Added rate limiting to AuthService"

1. autodoc_sync({ scope: 'outdated' })
   → Find outdated documents

2. autodoc_validate()
   → Check references

3. autodoc_save({
     filePath: 'src/services/auth/auth.service.md',
     content: '...'
   })

4. autodoc_changelog({ limit: 5 })
   → Show recent changes

4. Auto-Update (Watcher)

AutoDoc Watcher works automatically when enabled:

# ultrascript.yaml
mcp:
  autodoc:
    watcherEnabled: true
    debounceMs: 45000    # base delay

On .ts/.js file changes:

  1. Watcher detects changes via KnowledgeBus
  2. Debounce 30-60 sec (groups multiple changes)
  3. Updates AUTODOC.md in corresponding module:
    • Adds new exports
    • Removes deleted exports
    • Updates line numbers in references

Entity-Level Documentation Format

DON'T duplicate comments! Write about:

  • Role in system (WHY)
  • Who uses it (BY WHOM)
  • Participation in processes (WHERE)
  • Business invariants
  • Critical dependencies
  • Known issues
# AuthService

[→ auth.service.ts:15-120](auth.service.ts#L15-L120)

## Role in System

**Why needed**: Single point for authentication management...

## Who Uses

| Consumer | How |
|----------|-----|
| [→ API Gateway](../../gateway/README.md) | Token validation |

## Participation in Processes

- [→ FLOW.md#registration](../../FLOW.md#registration)
- [→ PROCESSES.md#auth-flow](../../PROCESSES.md#auth-flow)

## Business Invariants

- One active token per user
- Rate limiting: 5 attempts/min

Reference Syntax

[→ file.ts:25-50](file.ts#L25-L50)     # To code lines
[→ ModuleName](./_index.md)             # To documentation
[→ FLOW.md#scenario](../../FLOW.md#scenario)  # To section

In code comments:

/**
 * @see docs://.autodoc/PROCESSES.md#auth-flow
 * @flow user-registration, api-auth
 */

  • ultracode — Code analysis, semantic search, refactoring, code modification
  • ultracode-trace — Debugging, flow analysis, "why not called"

スコア

総合スコア

70/100

リポジトリの品質指標に基づく評価

SKILL.md

SKILL.mdファイルが含まれている

+20
LICENSE

ライセンスが設定されている

+10
説明文

100文字以上の説明がある

+10
人気

GitHub Stars 100以上

0/15
最近の活動

3ヶ月以内に更新がある

0/10
フォーク

10回以上フォークされている

0/5
Issue管理

オープンIssueが50未満

+5
言語

プログラミング言語が設定されている

+5
タグ

1つ以上のタグが設定されている

0/5

レビュー

💬

レビュー機能は近日公開予定です