LLM prompt update

Rewrote BABOK_AGENT/LLM_BABOK_AGENT/BABOK_Agent_LLM_Prompt.md for current capabilities:

• 9-stage pipeline (Stage 0 charter gate + Stages 1-8)
• 16 MCP tools and stage resources
• Plugin marketplace (Claude, Codex, Copilot)
• Slash commands with PL/ENG language selection
• CLI highlights, project storage model, quality checks

Install

/plugin marketplace update babok_analyst
/plugin install babok_analyst@babok_analyst
/reload-plugins

README update

Expanded plugin marketplace install guide with component table and troubleshooting matrix.

Install

/plugin marketplace add GSkuza/BABOK_ANALYST
/plugin marketplace update babok_analyst
/plugin install babok_analyst@babok_analyst
/reload-plugins

See README.md for full plugin documentation.

Plugin install fix

Claude Code plugin install no longer fails with agents: Invalid input. Manifest follows official Anthropic auto-discovery pattern.

Install

/plugin marketplace add GSkuza/BABOK_ANALYST
/plugin marketplace update babok_analyst
/plugin install babok_analyst@babok_analyst
/reload-plugins

Changes

• Fix .claude-plugin/plugin.json validation (remove invalid agents/commands/skills paths)
• Add Claude slash commands as commands/*.md
• Correct marketplace slug in docs + troubleshooting section
• All packages bumped to 2.2.0

Full notes: RELEASE_NOTES.md | CHANGELOG.md

Plugin Marketplace Install

Claude Code:

/plugin marketplace add GSkuza/babok_analyst
/plugin install babok_analyst@babok_analyst

Codex: codex plugin marketplace add GSkuza/babok_analyst

Copilot CLI:

copilot plugin marketplace add GSkuza/babok_analyst
copilot plugin install babok_analyst@babok_analyst

Highlights

• Plugin marketplace distribution (Claude / Codex / Copilot CLI)
• Portable MCP wiring — 16 tools + 9 stage resources
• Lifecycle hooks, bundled skills, agents, slash commands
• Per-stage LLM routing (deep analysis for stages 3, 4, 6, 8)
• Web UI (Next.js 15), quality scorer, validation engine, document ingestion
• Canonical project storage: projects/ (BABOK_Analysis/ = legacy CLI export only)

Full notes: RELEASE_NOTES.md | CHANGELOG.md

Key changes in the release:
a) llm.js — createLlmClient() stateless factory
b) engine.js — DEEP_ANALYSIS_STAGES = {3,4,6,8} + per-stage routing
c) run.js — --orchestrate path wires deepAnalysisClient
d) babok.js — new --deep-model CLI flag
CHANGELOG.md — 2.1.1 entry added

[2.1.0] - 2026-04-13

Added

🌐 Web UI (Next.js 15 App Router) — web/

• New web/ application with full Next.js 15.5 App Router setup.
• Dashboard — lists all projects with stage progress bars.
• New project form — name + language selection.
• Project detail view — stage pipeline with status indicators.
• Stage view — deliverable Markdown renderer + Approve/Reject buttons.
• Export page — one-click ZIP download of all deliverables.
• API routes: GET/POST /projects, GET /projects/[id], GET/POST /projects/[id]/stages/[n], GET /projects/[id]/export.
• Components: StageProgressBar, QualityScoreCard, DeliverableViewer, ApproveRejectButtons.
• lib/babok-client.ts — typed API client for server-side communication.

🤖 AI Reasoning Engine — cli/src/reasoning/

• Phase 1 — Quality Scorer (F3-T1):
  • cli/src/quality/scorer.js — scoreStage() / scoreAll() with rubric-based scoring: completeness (40%), SMART quality (30%), cross-stage consistency (30%).
  • cli/src/quality/checks/completeness.js — heading-regex section detector.
  • cli/src/quality/checks/smart.js — numeric / date / currency / ROI heuristics.
  • cli/src/quality/checks/consistency.js — intra-stage cross-reference checks.
  • New CLI command: babok score <id> <stage|all> with chalk score cards.
• Phase 1 — Cross-Stage Consistency Validator (F1-T3):
  • cli/src/validation/cross-stage-validator.js — validateProject() engine with 6 built-in rules:
    • FR Traceability (FR IDs → RTM coverage)
    • Budget Ceiling (Stage 8 vs Stage 1)
    • Integration Coverage (Stage 5 vs Stage 2)
    • KPI Coverage (Stage 2 vs Stage 1)
    • Critical Risk Owner (Stage 7)
    • Roadmap Date (Stage 6 vs Stage 1)
  • New CLI command: babok validate <id> — exits with code 1 on validation errors.
• Phase 2 — Multi-Agent Debate Pattern (F1-T1): adversarial reasoning loop for higher-quality stage outputs.
• Phase 2 — Chain-of-Verification (F1-T2): LLM self-verification pass on generated deliverables.
• Phase 2 — Gold Standard Evaluation Suite (F3-T2): automated benchmark comparison against reference outputs.

📥 Document Ingestion Pipeline (F2-T2) — cli/src/

• cli/src/lib/document-parser.js — parses PDF / DOCX / XLSX / CSV / TXT / MD files.
• cli/src/commands/ingest.js — new babok ingest <file> command with LLM-based document classification and tagging.
• cli/src/reasoning/prompts/ingest_tagger.md — LLM classification prompt.
• Ingested documents are listed in babok status output.

🗺️ Visual Process Mapping (F5-T2) — cli/src/

• cli/src/reasoning/process-mapper.js — generateProcessDiagram() with automatic retry.
• cli/src/reasoning/prompts/process_to_mermaid.md — LLM prompt generating Mermaid flowcharts.
• New --diagram flag for the babok run command (available for Stage 2 AS-IS and Stage 5 TO-BE).

🏗️ Architecture & Agent Configuration — docs/, BABOK_AGENT/

• docs/L2_L3_ARCHITECTURE.md — full L2/L3 agent layer design.
• docs/MCP_TOOLS_SPECIFICATION.md — MCP tools API reference.
• docs/MIGRATION_GUIDE_L1_to_L2.md — CLI-to-MCP migration guide.
• docs/workflows.md — end-to-end workflow diagrams.
• BABOK_AGENT/agents/ — multi-agent orchestration layer:
  • orchestrator_config.json, per-stage agent configs (stage1_config.json … stage8_config.json).
  • quality_audit_agent.md / quality_scoring_rubric.json.
  • context_schema_v2.json — updated context schema v2.
• BABOK_AGENT_SYSTEM_PROMPT.md — consolidated root-level system prompt.
• DEVELOPER_TASKS.md — full module breakdown with DoD and test plans for all 6 feature pillars.

📦 Document Templates — templates/

• BRD_Template.md, Risk_Register_Template.md, Stakeholder_Analysis_Template.md, User_Story_Template.md.
• project_context.example.json — reference context schema.

🧪 Test Suite (F6-T1) — tests/

• 73 tests passing (native node:test runner, ESM):
  • tests/unit/project.test.js (15 assertions)
  • tests/unit/journal.test.js (16 assertions)
  • tests/unit/scoring.test.js (14 assertions)
  • tests/unit/validation.test.js (18 assertions)
  • tests/integration/cli-workflow.test.js (10 steps)
• Fixture files: valid_stage1.md, invalid_stage1_missing_raci.md, valid_stage4.md, etc.
• Test helpers: mock-llm.js, temp-project.js.

📚 Knowledge Base — knowledge/

• 16 JSON benchmark / industry / regulatory / anti-pattern files.
• knowledge/README.md.

🛠️ Developer Tooling

• generate_manual.py — automated DOCX/PDF user manual generation from Markdown sources.
• .github/copilot-instructions.md — 1600-line Copilot Chat integration file.
• .github/prompts/ — stage-specific Copilot Chat prompt files (babok-stage-1.prompt.md … babok-run-all.prompt.md).
• .github/workflows/lint-prompts.yml — CI prompt linter on push/PR.

Changed

• Version bump: All packages synchronized to 2.1.0 (cli, babok-mcp, web, root workspace).
• babok-mcp/ context migrated to v2.0: Updated agent_config and stage configs to use new context_schema_v2.json.
• 6 new L2 MCP tools added to babok-mcp/: extending the MCP server API surface.

Fixed

• cli/src/lib/document-parser.js: Replaced vulnerable xlsx dependency with exceljs (security fix — xlsx had unpatched CVE).
• cli/src/lib/htmlToText(): Simplified tag-filter logic to resolve two CodeQL security warnings (tag injection path).
• web/: STAGE_NAMES constant corrected to use 1-based Record<number, string> (was 0-based, causing off-by-one display errors).

Security

• Removed xlsx@^0.18.5 from cli/package.json; replaced with exceljs@^4.4.0 — no known CVEs.

What's New in v2.0.2

1. Fixed setup.bat (Root Installer)

The root-level setup.bat was completely rewritten to fix multiple critical bugs:

• Removed UTF-8 box-drawing characters (─, �) that caused cmd.exe to misparse lines when combined with chcp 65001
• Added setlocal EnableDelayedExpansion so !errorlevel! works correctly inside if blocks (previously %errorlevel% was always 0 inside parentheses)
• Removed chcp 65001 — caused cmd.exe byte-counting mismatch leading to skipped/garbled output lines
• Replaced cmd /k at the end with pause >nul — no more orphaned nested shell after installation
• Fixed PATH refresh — reads fresh Machine+User registry PATH at startup; prevents "node not found" on fresh installs launched from Explorer
• Added Node.js fallback search — checks standard install locations (%ProgramFiles%, %LOCALAPPDATA%) if node isn't on PATH
• Fixed setx PATH safety — no longer passes the full expanded %PATH% (which would truncate at 1024 chars); now reads and extends only the User-scoped registry PATH

2. New babok-mcp/setup.bat (MCP Server Installer)

Added a dedicated one-click Windows installer for the babok-mcp server:

• Detects Node.js (with fallback location search)
• Runs npm install for MCP dependencies
• Validates server entry point with node --check (no hang — does not start the server)
• Displays the ready-to-paste JSON config block for claude_desktop_config.json, with dynamically resolved absolute paths (forward slashes, no .. segments)
• Offers to open claude_desktop_config.json in Notepad if Claude Desktop is installed
• Works in any install location (not hardcoded to D:)

3. Updated MCP User Manuals

• babok-mcp-podrecznik uzytkownika.md (PL) — updated to v2.0.2: added setup.bat as install option, described babok_rename_project and babok_delete_project tools (were missing), updated tool count from 8 to 10, corrected FAQ entries for rename/delete
• babok-mcp-user-manual.md (EN) — new file: full English translation of the MCP user manual, covering all 10 tools, 4 client configurations, troubleshooting, and FAQ

Upgrade Notes

• No breaking changes from v2.0.1.
• Existing projects remain fully compatible.
• setup.bat can be re-run safely on existing installations.

BABOK Analyst v1.8.2 - Release Notes

Release Date: February 11, 2026
Status: Minor Release
License: MIT

What's New in v1.8.2

1. Automated CLI Pipeline

• New babok run command to execute the analysis pipeline from a context file or short prompt.
• Supports stage selection (--stages 1,2,3), language override, output target, and --auto for fully automated runs.

2. Vertex AI Support

• Added Google Vertex AI as a provider with project/region configuration and optional service account credentials.
• Secure config storage for Vertex in the local encrypted keystore.

3. Improved Provider Flow

• Provider selection now includes model selection.
• Updated OpenAI and Hugging Face default models and model lists.
• babok chat now awaits provider initialization to avoid startup race conditions.

4. Prompt Helpers and Context Template

• Added .github/prompts/ for stage-specific and full-run prompt helpers.
• Added templates/project_context.example.json for standardized context input.

5. User Manual (PL)

• Added PDF user manual for distribution.

Upgrade Notes

• No breaking changes from v1.8.1.
• Existing projects remain compatible; new features are optional.

BABOK Analyst v1.8.1 - Release Notes

Release Date: February 8, 2026
Status: Hotfix Release
License: MIT

What's New in v1.8.1

Documentation & Versioning Hotfix

• Copilot Instructions Version Bump: Updated .github/copilot-instructions.md from v1.4 to v1.8.1 so that VS Code / Copilot Chat load the correct, up-to-date system prompt.
• Version Alignment: Synchronized VERSION, CLI cli/package.json, and top-level README.md footer to 1.8.1, matching the current feature set.

Upgrade Notes

• No behavioral changes to the agent logic compared to v1.8.0.
• Safe to update; existing projects and prompts remain fully compatible.

BABOK Analyst v1.8.0 - Release Notes

Release Date: February 8, 2026
Status: Stable Release
License: MIT

What's New in v1.8.0

1. Sequential Question Protocol

• Agent now asks questions one-by-one within each stage, showing STAGE N - QUESTION X/Y for clear progress.
• After each answer, the agent confirms with a short "âś… Answer recorded" summary.
• Before generating a document, the agent presents a consolidated summary of your answers for that step/stage.

2. Question Navigation Commands

• New high-level commands inside the conversation:
  • Next question – skip the current question and go to the next.
  • Previous question – go back within the same step.
  • Skip questions – show all remaining questions at once (batch mode).

3. CEO-Ready DOCX/PDF Exports

• New CLI commands:
  • babok make docx <project_id> – generate professionally formatted DOCX from stage .md files.
  • babok make pdf <project_id> – generate matching PDF exports.
  • babok make all <project_id> – create both DOCX and PDF in one run.
• Outputs are styled for executive presentation (headings, tables, headers/footers) and placed in <project_dir>/exports.

4. Documentation Updates

• Updated system prompts, Quick Start Guide, Command Reference, and main README to describe the sequential question flow and new export commands.

BABOK Analyst v1.7.0 - Release Notes

Release Date: February 8, 2026
Status: Stable Release
License: MIT

What's New in v1.7.0

Enhanced User Experience & Accessibility

1. Bilingual Quick Start Commands

• New commands: begin and zacznij as direct aliases for babok new
• Instant project creation with language-specific entry points
• Improved onboarding for English and Polish users
• Impact: Faster project initiation, reduced learning curve

2. LLM Management Enhancements

• Improved stability in provider switching
• Better error handling for API failures
• Enhanced connection management for 20+ supported models
• Impact: More reliable AI interactions across different providers

3. Project Management Improvements

• Enhanced project structure and journaling system
• Better reliability in project state management
• Improved multi-project handling
• Impact: More robust project lifecycle management

Technical Improvements

• Enhanced CLI entry points for better user experience
• Improved error messages and feedback
• Code quality and maintainability improvements

Upgrade Notes

• No breaking changes from v1.6.0
• All existing projects and configurations remain compatible
• New commands are additions, not replacements

Known Issues

• None at release time

Next Steps

See CHANGELOG.md for complete list of changes.

BABOK Analyst v1.1.0 - Release Notes

Release Date: February 7, 2026
Status: Stable Release
License: MIT
Reviewed By: Grzegorz Skuza (GTMO Framework Author, AI Safety Specialist)

What's New in v1.1.0

8 Critical Improvements Based on Expert Review

1. Short Rationale + Evidence (replaces Chain-of-Thought) - CRITICAL

• Every conclusion now includes: clear statement, key assumptions (max 3-5), evidence source
• Internal reasoning process no longer exposed (security + usability improvement)
• Impact: ~60% reduction in output verbosity, improved readability

2. Executive Summary (1 page per stage) - CRITICAL

• Added to all stage deliverable templates
• Includes: key findings, critical decisions, business impact, approval requirements
• Agent presents summary FIRST before detailed analysis
• Impact: Stakeholders actually read summaries; approval becomes informed decision

3. Change Control Process - CRITICAL

• New Section 9 in Stage 4 with formal CR process
• Change Request template, Impact Analysis checklist, Approval Matrix
• Requirements versioning (semantic), baseline freeze rules
• Impact: Prevents scope creep, maintains audit trail

4. Mid-Market Positioning (replaces SME/MSP)

• Updated to: Manufacturing, Distribution, Service Industries
• Company profile: €10-100M revenue, 50-500 employees
• EU/Polish regulatory focus (GDPR, KSeF, sector-specific)

5. DPIA as Explicit Deliverable

• GDPR Article 35 compliance template in Stage 7
• Processing overview, necessity assessment, risk matrix, mitigation measures
• Data subject rights implementation plan with SLAs

6. KSeF Technical Requirements Expansion

• FR-020 expanded from basic to 9 detailed acceptance criteria
• Covers: normal flow, validation errors, retry logic, duplicate prevention
• Monitoring dashboard, environment management, authentication, edge cases

7. RACI Matrix

• Added to Stage 1 (Stakeholder Mapping)
• 10 key project activities with clear R/A/C/I assignments
• Steering Committee structure and escalation path

8. Modeling Notation Standards

• BPMN 2.0, UML 2.5, C4 Model, VSM standards
• Quality checklist for all diagrams
• Added to Stage 2 (applies to Stage 5 as well)

BABOK Analyst v1.0.0 - Release Notes (Previous)

Release Date: February 7, 2026
Status: Superseded by v1.1.0
License: MIT

Welcome to BABOK Analyst 1.0.0!

This is the initial public release of BABOK Analyst - an AI-powered business analysis agent that implements the BABOK v3 framework for professional bus...

What's New in v2.0.1

🚀 One-Click Setup for Non-Technical Analysts

Running setup.bat (Windows) or setup.sh (Linux/macOS) now handles the entire installation:

• Checks for Node.js (with download link if missing)
• Runs npm install automatically
• Optionally adds babok to PATH
• Launches the interactive babok setup wizard

The babok setup (alias: babok init) wizard guides the analyst through:

1. Language selection (EN / PL)
2. AI provider selection with direct links to API key pages
3. API key entry + live connectivity test
4. Optional first project creation — no JSON editing required

✏️ Full Project CRUD

Two new CLI commands:

Also available in the MCP server as:

• babok_rename_project (tool 9)
• babok_delete_project (tool 10, requires confirm_id matching the project ID)

🔒 File Locking for Team Collaboration

When multiple analysts share a project directory (network drive, synced folder):

• babok chat and babok run now acquire a lock before editing a stage
• Lock file: <project_dir>/.stage_N.lock — contains locked_by, hostname, pid, locked_at
• A second user trying to open the same stage sees:

  ⛔ Stage 3 is currently locked by another user:
     anna@WORKSTATION-02 (PID 14872), locked 12 min ago
  

• Locks older than 2 hours are automatically removed (stale detection)
• Lock released cleanly on /exit, Ctrl+C, or stage completion

📝 Updated README

README updated to v2.0.1 with:

• New Quick Start section with setup.bat / setup.sh
• Full command table including setup, rename, delete
• Team Collaboration & File Locking section
• 10-tool MCP table
• 5-provider chat table (added Vertex AI)

Files Changed

• setup.bat — Windows installer (new)
• setup.sh — Linux/macOS installer (new)
• cli/src/commands/setup.js — setup wizard (new)
• cli/src/commands/rename.js — rename command (new)
• cli/src/commands/delete.js — delete command (new)
• cli/src/lock.js — file locking module (new)
• cli/src/commands/chat.js — integrated locking
• cli/src/commands/run.js — integrated locking
• cli/bin/babok.js — registered new commands, bumped to 2.0.1
• babok-mcp/src/server.js — added tools 9 & 10
• cli/package.json, babok-mcp/package.json — version 2.0.1
• README.md — fully updated

v2.0.0 — MCP Server for BABOK Agent

What's new

This release adds babok-mcp — a Model Context Protocol server that exposes BABOK project management tools to Claude Desktop, Claude Code, Cursor, VS Code Copilot, and any MCP-compatible AI assistant.

No more switching between terminal and chat. Claude can now create projects, load stage instructions, approve stages, search deliverables, and export results directly via tool calls.

MCP Tools (8)

MCP Resources (9)

All stage prompt files exposed as babok://stages/{n} URIs — Stage 0 through Stage 8.

Quick setup for Claude Desktop

cd BABOK_ANALYST/babok-mcp
npm install

Add to claude_desktop_config.json:

{
  "mcpServers": {
    "babok": {
      "command": "node",
      "args": ["D:/BABOK_ANALYST/babok-mcp/bin/babok-mcp.js"],
      "env": {
        "BABOK_PROJECTS_DIR": "D:/BABOK_ANALYST/projects",
        "BABOK_AGENT_DIR":    "D:/BABOK_ANALYST/BABOK_AGENT/stages"
      }
    }
  }
}

Restart Claude Desktop. A tools icon confirms the server is connected.

Test results

10/10 smoke tests passed
- generateProjectId format OK
- createJournal: stages 0-8 initialized correctly
- listProjectIds: project appears in directory scan
- resolveProjectId: partial suffix "4LGM" resolves to full ID
- readJournal: reads JSON correctly
- approveStage(0): Stage 0 approved, Stage 1 set to in_progress
- double-approve throws "already approved"
- rejectStage: status set to rejected, notes recorded
- getDeliverable(missing): returns null
- getDeliverable(saved file): returns full content

Files changed

• babok-mcp/ — new package (MCP server)
• CHANGELOG.md — v2.0.0 entry
• README.md — MCP section added, architecture diagram updated
• VERSION — 2.0.0