v0.5.4

Changelog

[0.5.4] - 2026-05-01

Added

• (Add your changes here)

Changed

• (Add your changes here)

Fixed

• (Add your changes here)

[0.5.3] - 2026-05-01

Added

• docuflow sync --ai — One-shot wiki sync (AI-powered locally)
  • Auto-detects best AI bridge (Copilot → Claude Code → Codex)
  • --copilot, --claude, --codex flags to force bridge
  • --since-commit REF for incremental sync
  • --fail-on-score N to fail if health < threshold
  • --quiet mode for scripting/CI integration
• Local-first development — Git post-commit hook auto-syncs on every commit
  • No GitHub Actions workflow overhead
  • No API key management needed
  • Instant local feedback on wiki health
• Comprehensive setup guide — LOCAL_SYNC_SETUP.md with workflows and troubleshooting

Changed

• CLI now prioritizes local AI bridges for faster, more reliable sync
  • Copilot CLI (MCP direct calling) preferred
  • Claude Code and Codex supported
  • ANTHROPIC_API_KEY fallback for API-based sync

Fixed

• Watch daemon graceful recovery on AI bridge timeouts
• Proper handling of no-code-changes commits

[0.5.2] - 2026-05-01

Added

• (Add your changes here)

Changed

• (Add your changes here)

Fixed

• (Add your changes here)

[0.5.1] - 2026-05-01

Added

• (Add your changes here)

Changed

• (Add your changes here)

Fixed

• (Add your changes here)

[0.5.0] - 2026-05-01

Added

docuflow watch — Auto-sync daemon

• Background daemon that watches for changes and syncs the wiki automatically
• Source file watcher: drop a .md into .docuflow/sources/ and it's ingested in <1 second
• Code file watcher: detects changes to .ts, .py, .go, .rb, .java, .cs etc. and triggers AI-powered documentation
• Scheduled lint: runs lint_wiki every N hours (default: 24h) and reports health score
• --lint-interval N, --code-ext ts,py flags
• --ai flag enables the AI bridge (auto-detects best available)
• --copilot, --claude, --codex flags to force a specific AI bridge

docuflow watch stop/status/restart — Daemon lifecycle

• watch stop — gracefully stop the running daemon (SIGTERM → 5s wait → SIGKILL if needed)
• watch status — see if daemon is running: shows ● running, PID, uptime, bridge, started time
• watch restart — stop + restart with identical options automatically
• Auto-cleans stale PID files if the process died unexpectedly

docuflow sync — One-shot sync for CI/CD and git hooks

• Re-ingest all sources, rebuild index, run health check — in one command
• --source <file> — sync a single file
• --no-lint — skip health check (faster)
• --fail-on-score N — exit 1 if health score < N (CI quality gate, default: 70)
• --quiet — suppress output for clean CI logs
• --since-commit <REF> — only process code that changed since a git ref
• --ai — AI-powered sync: detects changed code and auto-documents it

AI bridge — 4 supported AI engines

Priority	Bridge	How it syncs
1	@github/copilot CLI	Directly calls DocuFlow MCP tools (ingest, index, lint) ⚡
2	claude CLI (Claude Code)	Directly calls DocuFlow MCP tools ⚡
3	codex CLI (OpenAI Codex)	Generates doc text → saves to sources/ → ingests
4	ANTHROPIC_API_KEY	Same as codex via direct HTTPS API

Key insight: When @github/copilot or claude is used, the AI agent directly calls DocuFlow MCP tools (ingest_source, update_index, lint_wiki) and returns a full wiki maintenance report — no intermediate step needed.

Git hook auto-installation

• docuflow init now installs .git/hooks/post-commit automatically
• After every git commit, the wiki syncs in the background (never delays your git workflow)
• Uses the best available AI bridge automatically

Fixed

• Double-fire on macOS fs.watch — debounce prevents duplicate ingestion
• setInterval 32-bit overflow for large lint intervals
• Claude CLI bridge now passes --dangerously-skip-permissions for non-interactive MCP tool use

Added

• docuflow init now registers DocuFlow in OpenAI Codex CLI (~/.codex/config.toml) — MCP tools available in every Codex session automatically
• docuflow init now generates AGENTS.md in the project root — Codex reads DocuFlow tool instructions automatically (mirrors CLAUDE.md for Claude Code)
• docuflow init output now lists all 5 targets: Claude Desktop, VS Code Copilot (user), Copilot CLI, Codex CLI, Workspace

Added

• docuflow init now registers DocuFlow in GitHub Copilot CLI (~/.copilot/mcp-config.json) — tools available in gh copilot agent sessions automatically
• docuflow init now writes a project-level .vscode/mcp.json — commit it to share DocuFlow with your whole team without each member needing to run init
• docuflow init output now shows all 4 registration targets: Claude Desktop ✓, VS Code Copilot (user) ✓, Copilot CLI ✓, Workspace ✓

[0.4.4] - 2026-04-23

Added

• (Add your changes here)

Changed

• (Add your changes here)

Fixed

• (Add your changes here)

[0.4.3] - 2026-04-23

Added

• docuflow init now registers DocuFlow in GitHub Copilot (VS Code user MCP config) in addition to Claude Desktop — one command sets up both AI tools
• docuflow --version / -v flag to print installed version

[0.4.2] - 2026-04-23

Added

• docuflow --version and docuflow -v — print the installed version
• Improved bare docuflow help output with version header and structured sections

[0.4.1] - 2026-04-23

Added

• (Add your changes here)

Changed

• (Add your changes here)

Fixed

• (Add your changes here)

[0.4.0] - 2026-04-23

Added

• generate_dependency_graph — 15th MCP tool. Scans a project and builds an import/shared-table/shared-endpoint graph. Returns nodes, edges, shared_tables, shared_endpoints, and most_connected (top 10 highest-risk files). Supports focus (neighbourhood filter) and extensions filter.
• docuflow suggest — New CLI command. Domain-aware first-steps guidance — auto-detects your domain (Code/Research/Business/Personal) from .docuflow/schema.md, prints 5 prioritised wiki page suggestions with reasons and ready-to-paste Claude prompts.
• CLAUDE.md generation — docuflow init and docuflow init --interactive now write CLAUDE.md at the project root. Contains all 15 tool descriptions, common workflows, and storage layout. Idempotent: safe to run multiple times.
• Staleness detection — list_wiki returns stale: boolean per page and stale_pages total count. read_specs returns stale: boolean per spec. Threshold: 30 days since last update.
• Go extraction — Struct/interface types, func declarations, import blocks, os.Getenv, gorilla/mux/gin/chi/echo HTTP routes, GORM table references.
• Ruby/Rails extraction — Class/module/def declarations, require, ENV[], Rails route helpers (get, post, resources), ActiveRecord associations and explicit table names.
• Enhanced docuflow status — Now shows package version, CLAUDE.md presence, wiki page counts by category, source file count, last ingest date, and smart hints.
• Richer ingest_source pages — Entity and concept pages now include the surrounding paragraph from the source document instead of an empty "Introduced in" stub.
• Dynamic preview_generation — Previews now read actual wiki page count and source file size before producing estimates instead of using hardcoded strings.

Fixed

• lint_wiki path bug — All health check functions were looking for pages at wiki/pageId.md (flat) instead of wiki/entities/pageId.md (subdirectory). Result: every lint check silently returned 0 issues. Now correctly resolves full file paths.
• Category pluralization bug — "entities".replace("s","") → "entitie" (not "entity"). Fixed in list_wiki, wiki_search, update_index, and save_answer_as_page using a lookup map.
• save_answer_as_page links — Related Pages linked to ../CATEGORY/pageId.md with the literal string "CATEGORY". Now resolves the actual directory name.
• list_wiki filter bug — Filtering by category entity was building path wiki/entitys/. Fixed with SINGULAR_TO_PLURAL map.
• init-interactive.ts tip — Misleading "open .claude/instructions.md" tip now correctly references CLAUDE.md.

Changed

• Tool count: 14 → 15
• CLI commands: 2 → 3 (suggest added)

[0.2.0] - 2026-04-16

Added

LLM Wiki Pattern Implementation (Phases 1-5)

• Complete LLM Wiki architecture: persistent, incrementally-maintained knowledge bases
• 12 MCP tools for wiki management:
  • Ingest & Index: ingest_source, update_index, list_wiki
  • Query & Synthesis: wiki_search, query_wiki, answer_synthesis, save_answer_as_page
  • Maintenance: lint_wiki with health scoring, contradiction detection, orphan page detection
• Domain-specific wiki schemas (4 templates: Code, Research, Business, Personal)
• Real-world example: Docuflow's own 188-page wiki at 100% health score

User Experience Enhancement (Phase 6)

• Copilot Auto-Discovery: .claude/instructions.md (35 KB) teaches Claude about Docuflow

  • Claude reads at session start and auto-discovers when to use Docuflow
  • No longer requires explicit instruction to call Docuflow tools

• Tool Transparency: Two new guidance tools

  • preview_generation — Shows what tools will do before running (predicted actions, impact level, files affected)
  • get_schema_guidance — Recommends what documents should exist based on domain and wiki state

• Comprehensive Documentation (110+ KB)

  • TROUBLESHOOTING.md — Problem-solving guide (command not found, MCP issues, wiki quality, performance, data safety)
  • WHEN_TO_USE.md — Decision framework with matrix, cost-benefit analysis, domain-specific guidance
  • COPILOT_INTEGRATION.md — LLM agent integration reference
  • USAGE_EXAMPLES.md — 6 real-world workflows
  • BEST_PRACTICES.md — Maintenance guidelines
  • EXAMPLE_SCHEMAS.md — 4 domain-specific templates
  • LLM_WIKI_PATTERN.md — Deep dive on pattern philosophy

• Interactive Initialization

  • docuflow init --interactive — Domain-aware setup (Code/Research/Business/Personal)
  • Guided prompts for project info
  • Auto-generated domain-specific schema
  • Planning template with first sources and questions
  • Next steps guidance

Changed

• Updated all 12 MCP tools to support domain-specific schemas
• Enhanced README with Phase 6 info, 14 tools, and getting started guide
• Updated monorepo documentation with new tool locations

Fixed

Resolved All Pre-LLM-Wiki Testing Concerns:

1. ✅ Claude doesn't auto-discover Docuflow → .claude/instructions.md
2. ✅ Tool execution feels like a "black box" → preview_generation tool
3. ✅ No guidance on document planning → get_schema_guidance + decision frameworks
4. ✅ Poor onboarding → Interactive init with domain templates
5. ✅ No troubleshooting → TROUBLESHOOTING.md guide
6. ✅ Unclear decision-making → WHEN_TO_USE.md matrix
7. ✅ No transparency in tool behavior → Preview tool + clear predictions

Quality Metrics

• ✅ 127+ tests passing (100% pass rate across all 6 phases)
• ✅ 0 breaking changes (100% backward compatible)
• ✅ 14 MCP tools (4 legacy + 10 new LLM Wiki tools)
• ✅ Build clean (0 errors, 0 warnings)
• ✅ Real-world validation (Docuflow's own 188-page wiki)

Migration & Breaking Changes

None. This release is fully backward compatible with 0.1.x.

All legacy tools (read_module, list_modules, write_spec, read_specs) continue to work unchanged.

Documentation

• New User Guide: Start with docs/WHEN_TO_USE.md to decide if Docuflow is right for you
• Getting Started: Run docuflow init --interactive for domain-aware setup
• For Claude Users: LLM agents auto-discover Docuflow via .claude/instructions.md
• Troubleshooting: See docs/TROUBLESHOOTING.md for common issues and solutions
• Examples: See docs/USAGE_EXAMPLES.md for real-world workflows

---

All notable changes to Docuflow are documented here.
Format follows Keep a Changelog
Versioning follows Semantic Versioning

---

[0.1.1] — 2026-04-09

Added

• Package README files on npmjs.com for both @doquflow/server and @doquflow/cli

---

[0.1.0] — 2026-04-09

First public release

Core MCP Tools

• read_module — Read one source file, extract classes, functions, dependencies,
  DB tables, endpoints, config refs, raw content (truncated 8 000 chars)
• list_modules — Walk a project directory, bulk-extract all non-binary files,
  return structured facts per file (no raw content for performance)
• write_spec — Write a markdown spec to .docuflow/specs/<name>.md,
  update the per-project index. Serialised per project to prevent race conditions.
• read_specs — Read saved specs back; optionally filter by module name

Language Support

TypeScript, JavaScript, Python, Go, Rust, Java, C#, PHP, Ruby, Kotlin, Swift,
Angular, Vue, HTML, SQL, Shell, PowerShell, YAML, JSON, and more.

Extraction Engine

Regex-based extraction for classes, interfaces, functions, dependencies,
database tables (SQL + Entity Framework), REST endpoints (.NET, Express, NestJS),
and config/environment references.

Developer Experience

• npx @doquflow/cli init — registers Docuflow MCP in Claude Desktop config
• npx @doquflow/cli status — shows spec count and registration state
• Zero AI calls inside the server — all intelligence stays in the agent
• No API keys, no network calls, no AI dependencies