ruvnet-kb-first 5.0.0 → 6.0.0

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (4) hide show
  1. package/README.md +41 -23
  2. package/SKILL.md +153 -75
  3. package/package.json +1 -1
  4. package/src/mcp-server.js +592 -258
package/README.md CHANGED
@@ -1,9 +1,9 @@
1
- Updated: 2026-01-02 09:45:00 EST | Version 5.0.0
1
+ Updated: 2026-01-02 10:20:00 EST | Version 6.0.0
2
2
  Created: 2026-01-01 15:28:53 EST
3
3
 
4
- # RuvNet KB-First Application Builder v5.0
4
+ # RuvNet KB-First Application Builder v6.0
5
5
 
6
- ## Build Intelligent Applications on Expert Knowledge
6
+ ## Score-Driven Architecture: Measurement IS Enforcement
7
7
 
8
8
  [![npm version](https://img.shields.io/npm/v/ruvnet-kb-first.svg)](https://www.npmjs.com/package/ruvnet-kb-first)
9
9
  [![License: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](https://opensource.org/licenses/MIT)
@@ -59,42 +59,60 @@ This is **not just RAG** (Retrieval-Augmented Generation). RAG retrieves context
59
59
 
60
60
  ---
61
61
 
62
- ## 🚀 Quick Start
62
+ ## 🚀 Quick Start (v6.0 - Score-Driven)
63
63
 
64
- ### Option 1: NPX (Recommended)
64
+ ### MCP Server (Recommended)
65
65
 
66
- ```bash
67
- # Initialize KB-First in any project
68
- npx ruvnet-kb-first init
66
+ Add to your Claude Code `settings.json`:
69
67
 
70
- # Check project status
71
- npx ruvnet-kb-first status
68
+ ```json
69
+ {
70
+ "mcpServers": {
71
+ "ruvnet-kb-first": {
72
+ "command": "npx",
73
+ "args": ["ruvnet-kb-first", "mcp"]
74
+ }
75
+ }
76
+ }
77
+ ```
78
+
79
+ Then use the 4 MCP tools:
72
80
 
73
- # Run compliance score
74
- npx ruvnet-kb-first score --detailed
81
+ ```
82
+ kb_first_assess # ALWAYS FIRST - establish baseline score
83
+ kb_first_phase 0 # Get phase guidance (shows baseline reminder)
84
+ # ... do the work ...
85
+ kb_first_delta # Compare scores - BLOCKS if regression
86
+ kb_first_gate 0 # Pass gate, proceed to next phase
75
87
  ```
76
88
 
77
- ### Option 2: Global Install
89
+ ### NPX (CLI Wrapper)
78
90
 
79
91
  ```bash
80
- npm install -g ruvnet-kb-first
92
+ # Start MCP server
93
+ npx ruvnet-kb-first mcp
81
94
 
82
- ruvnet-kb-first init
83
- ruvnet-kb-first status
84
- ruvnet-kb-first score
95
+ # Legacy CLI commands still work
96
+ npx ruvnet-kb-first score
97
+ npx ruvnet-kb-first status
85
98
  ```
86
99
 
87
- ### Option 3: Claude Code Slash Command
100
+ ### The Score-Driven Workflow
88
101
 
89
102
  ```
90
- /kb-first # Start interactive builder
91
- /kb-first init # Initialize in current project
92
- /kb-first score # Score KB compliance
103
+ ┌──────────────────────────────────────────────────────────────┐
104
+ │ 1. kb_first_assess → Baseline: 47/100 (Grade: D) │
105
+ │ 2. kb_first_phase → Phase 3: KB Population tasks │
106
+ │ 3. [Do the work] → Add content, generate embeddings │
107
+ │ 4. kb_first_delta → Current: 62/100, Delta: +15 ✓ PASS │
108
+ │ 5. kb_first_gate → Gate passed, proceed to Phase 4 │
109
+ │ 6. REPEAT → New baseline: 62/100 │
110
+ └──────────────────────────────────────────────────────────────┘
93
111
  ```
94
112
 
95
- ### Option 4: MCP Server (Deep Integration)
113
+ **Key Rule:** Negative delta BLOCKS progress. No shortcuts.
96
114
 
97
- Add to your Claude Code `settings.json`:
115
+ ### Legacy Options
98
116
 
99
117
  ```json
100
118
  {
package/SKILL.md CHANGED
@@ -1,42 +1,48 @@
1
- Updated: 2026-01-02 09:45:00 EST | Version 5.0.0
1
+ Updated: 2026-01-02 10:15:00 EST | Version 6.0.0
2
2
  Created: 2026-01-01 15:00:00 EST
3
3
 
4
- # RuvNet KB-First Application Builder v5.0
4
+ # RuvNet KB-First Application Builder v6.0
5
5
 
6
- ## Intelligence-First Architecture for Knowledge Applications
6
+ ## Score-Driven Architecture: Scoring IS Enforcement
7
7
 
8
- **Version:** 5.0.0
8
+ **Version:** 6.0.0
9
9
  **NPM Package:** `ruvnet-kb-first`
10
- **Purpose:** Build applications where curated expert knowledge drives intelligent decision-making through GNN reasoning, attention routing, and SONA learning.
10
+ **Philosophy:** Every operation requires baseline scoring. Every change shows delta. Negative delta BLOCKS progress. No shortcuts.
11
11
 
12
12
  ---
13
13
 
14
- ## What's New in v5.0
14
+ ## What's New in v6.0 - BREAKING CHANGES
15
15
 
16
16
  | Feature | Description |
17
17
  |---------|-------------|
18
- | **NPM Package** | `npx ruvnet-kb-first init` - works in any repo |
19
- | **CLI Commands** | `kb-first init`, `score`, `verify`, `status`, `phase`, `hooks` |
20
- | **MCP Server** | Deep Claude Code integration via `mcp-server.js` |
21
- | **Auto-Detection** | Automatic KB-First project detection on session start |
22
- | **12 Phases** | Complete build process (0-11 + 1.5) with 57 sub-phases |
23
- | **Dual Binary** | Both `ruvnet-kb-first` and `kb-first` work |
18
+ | **Score-Driven Architecture** | Scoring is THE enforcement mechanism, not just a metric |
19
+ | **4 MCP Tools** | Simplified from 7 CLI commands to 4 focused MCP tools |
20
+ | **Delta Enforcement** | Every phase requires before/after comparison |
21
+ | **Hard Gate Blocking** | Negative score delta BLOCKS progress - no bypass |
22
+ | **Baseline Requirement** | Cannot start work without establishing baseline |
23
+ | **Automatic Refresh** | Baseline resets after each gate passage |
24
24
 
25
- ### Quick Start
25
+ ### The 4 MCP Tools
26
26
 
27
- ```bash
28
- # Option 1: NPX (recommended)
29
- npx ruvnet-kb-first init
27
+ | Tool | Purpose | When to Use |
28
+ |------|---------|-------------|
29
+ | `kb_first_assess` | Calculate baseline scores | **FIRST** - before any work |
30
+ | `kb_first_phase` | Get phase guidance with baseline | After assess, shows tasks |
31
+ | `kb_first_delta` | Compare current vs baseline | After work, shows improvement |
32
+ | `kb_first_gate` | Hard gate - blocks on regression | When ready to proceed |
30
33
 
31
- # Option 2: Global install
32
- npm install -g ruvnet-kb-first
33
- kb-first init
34
+ ### The Score-Driven Workflow
34
35
 
35
- # Option 3: Slash command (in Claude Code)
36
- /kb-first
36
+ ```
37
+ 1. kb_first_assess → Establishes baseline (saves to .ruvector/baseline.json)
38
+ 2. kb_first_phase → Get phase tasks, see baseline reminder
39
+ 3. [Do the work]
40
+ 4. kb_first_delta → Compare scores, PASS if delta ≥ 0, FAIL if < 0
41
+ 5. kb_first_gate → Verify gate, proceed OR blocked
42
+ 6. REPEAT from step 1 for next phase
37
43
  ```
38
44
 
39
- ### MCP Server Setup
45
+ ### Quick Start (MCP Only)
40
46
 
41
47
  Add to your Claude Code settings:
42
48
 
@@ -51,71 +57,143 @@ Add to your Claude Code settings:
51
57
  }
52
58
  ```
53
59
 
60
+ Then use the tools:
61
+ ```
62
+ kb_first_assess # Start here - get your baseline
63
+ kb_first_phase 0 # Get Phase 0 (Assessment) guidance
64
+ # ... do the work ...
65
+ kb_first_delta # Check your improvement
66
+ kb_first_gate 0 # Pass gate, move to Phase 1
67
+ ```
68
+
54
69
  ---
55
70
 
56
- ## What's New in v4.3
71
+ ## Scoring Formula (100 Points Total)
72
+
73
+ ### KB Score (40 points)
74
+ | Component | Points | Measurement |
75
+ |-----------|--------|-------------|
76
+ | Entries | 10 | KB has content (2 pts per 5 entries) |
77
+ | Coverage | 10 | Domain docs exist (2 pts per doc) |
78
+ | Embeddings | 10 | Vectors generated |
79
+ | Freshness | 10 | Updated within 24h=10, 7d=7, 30d=4 |
80
+
81
+ ### App Score (40 points)
82
+ | Component | Points | Measurement |
83
+ |-----------|--------|-------------|
84
+ | KB Citations | 15 | Code files cite KB sources |
85
+ | Gap Resolution | 10 | 10 - (unresolved gaps) |
86
+ | Test Coverage | 10 | Tests exist and pass |
87
+ | Security | 5 | .gitignore excludes .env |
88
+
89
+ ### Process Score (20 points)
90
+ | Component | Points | Measurement |
91
+ |-----------|--------|-------------|
92
+ | Phase Completion | 10 | % of phases completed |
93
+ | Gates Passed | 5 | % of gates verified |
94
+ | Documentation | 5 | README, API docs, architecture |
57
95
 
58
- | Feature | Description |
59
- |---------|-------------|
60
- | **Phase 9: Security Audit** | OWASP Top 10, SQL injection, secrets scanning |
61
- | **Phase 10: Documentation** | Complete docs, API specs, versioning |
62
- | **Phase 11: Deployment** | Production deployment with public access |
63
- | **Security Scripts** | Automated security verification (9-security-audit.sh) |
64
- | **Go-Live Checklist** | Pre-launch and post-launch verification |
96
+ ---
65
97
 
66
- ## What's New in v4.2
98
+ ## How It Works
67
99
 
68
- | Feature | Description |
69
- |---------|-------------|
70
- | **Phase 1.5: Hooks Setup** | Automatic KB enforcement via RuVector hooks |
71
- | **Hook Pre-training** | ReasoningBank seeded with KB-First patterns |
72
- | **Hook Verification** | Comprehensive verification script (1.5-hooks-verify.sh) |
73
- | **Global Skill Install** | Install to ~/.claude/skills/ for all repos |
74
- | **Init Script** | One-command project initialization |
100
+ ### 1. `kb_first_assess` - Establish Baseline
75
101
 
76
- ## What's New in v4.1
102
+ **ALWAYS RUN FIRST.** Returns:
103
+ ```json
104
+ {
105
+ "action": "BASELINE_ESTABLISHED",
106
+ "total": 47,
107
+ "grade": "D",
108
+ "summary": { "kb": "12/40", "app": "25/40", "process": "10/20" },
109
+ "breakdown": { ... component details ... }
110
+ }
111
+ ```
77
112
 
78
- | Feature | Description |
79
- |---------|-------------|
80
- | **Greenfield Support** | Full workflow for new projects via PROJECT_INTENTIONS.md |
81
- | **Interactive Discovery** | Structured questions when no intentions file exists |
82
- | **IS/SHOULD/COULD Analysis** | Present vision to user before execution |
83
- | **User Feedback Loop** | Mandatory feedback before proceeding |
84
- | **CI/CD Pipeline** | GitHub Actions workflow for quality gates |
85
- | **Verification Scripts** | 8 executable scripts (8.1-8.8) for Phase 8 |
113
+ ### 2. `kb_first_phase` - Get Phase Guidance
86
114
 
87
- ## What's in v4.0
115
+ **Shows phase tasks and reminds you of baseline:**
116
+ ```json
117
+ {
118
+ "phase": 3,
119
+ "name": "KB Population",
120
+ "baseline": { "score": 47, "grade": "D" },
121
+ "tasks": ["Collect domain content", "Generate embeddings", ...],
122
+ "reminder": "⚠️ Run kb_first_delta when complete"
123
+ }
124
+ ```
88
125
 
89
- | Feature | Description |
90
- |---------|-------------|
91
- | **Phase 0: Assessment** | Mandatory scoring before any changes |
92
- | **Hard Gates** | DO NOT PROCEED enforcement at every phase boundary |
93
- | **Scoring Formulas** | Explicit 0-100 scoring for KB quality and app compliance |
94
- | **Swarm Integration** | Parallel execution configuration for all phases |
95
- | **Delta Tracking** | Before/after comparison with transformation report |
96
- | **Phase 8 Sub-Phases** | 8 numbered verification checks (8.1-8.8) |
126
+ ### 3. `kb_first_delta` - Measure Improvement
127
+
128
+ **THE ENFORCEMENT MECHANISM:**
129
+ ```json
130
+ {
131
+ "verdict": "PASS",
132
+ "canProceed": true,
133
+ "baseline": { "score": 47 },
134
+ "current": { "score": 62 },
135
+ "delta": { "total": "+15", "kb": "+10", "app": "+3", "process": "+2" }
136
+ }
137
+ ```
138
+
139
+ Or if regression:
140
+ ```json
141
+ {
142
+ "verdict": "FAIL",
143
+ "canProceed": false,
144
+ "delta": { "total": "-3" },
145
+ "blockReason": "Score dropped by 3 points. You CANNOT proceed until score improves."
146
+ }
147
+ ```
148
+
149
+ ### 4. `kb_first_gate` - Pass or Block
150
+
151
+ **Hard gate that enforces progress:**
152
+ ```json
153
+ {
154
+ "canProceed": true,
155
+ "gateStatus": "PASSED",
156
+ "scoreImprovement": "+15",
157
+ "nextPhase": 4,
158
+ "nextPhaseName": "Scoring & Gaps"
159
+ }
160
+ ```
161
+
162
+ Or if blocked:
163
+ ```json
164
+ {
165
+ "canProceed": false,
166
+ "blockReason": "GATE_BLOCKED: Score regression detected (-3 points)."
167
+ }
168
+ ```
169
+
170
+ ---
171
+
172
+ ## Why Score-Driven?
173
+
174
+ > "When you just stare at something, you take shortcuts."
175
+
176
+ The previous version (v5.0) had 7 CLI commands, a skill file, a command file, and an MCP server. **Too much complexity, not enough rigor.**
177
+
178
+ v6.0 simplifies to:
179
+ - **One interface:** MCP server with 4 tools
180
+ - **One enforcement mechanism:** Scoring with delta comparison
181
+ - **One rule:** No negative deltas allowed
182
+
183
+ This prevents:
184
+ - Skipping phases
185
+ - Claiming "good enough" without measurement
186
+ - Forgetting to verify improvements
187
+ - Proceeding with regressions
97
188
 
98
189
  ---
99
190
 
100
- ## How to Use This Skill
101
-
102
- When you invoke this skill, Claude will:
103
-
104
- 1. **Detect** application type (greenfield vs brownfield)
105
- 2. **Score** existing KB and app compliance (brownfield only)
106
- 3. **Confirm** transformation scope with user (brownfield only)
107
- 4. **Identify** which intelligence pattern fits
108
- 5. **Execute** the 9-phase build process with HARD quality gates
109
- 6. **Enforce** KB-First rules throughout development
110
- 7. **Verify** all 8 verification sub-phases pass
111
- 8. **Report** delta (before/after scores)
112
-
113
- **Trigger phrases:**
114
- - "Build [description] with KB-First" (greenfield)
115
- - "Apply KB-First to my existing app" (brownfield)
116
- - "Score my existing KB"
117
- - "What's my app compliance score?"
118
- - "/kb-first" (invoke skill directly)
191
+ ## Trigger Phrases
192
+
193
+ - "Score my project" `kb_first_assess`
194
+ - "What phase am I on?" → `kb_first_phase`
195
+ - "Did I improve?" `kb_first_delta`
196
+ - "Can I proceed?" `kb_first_gate`
119
197
 
120
198
  ---
121
199
 
package/package.json CHANGED
@@ -1,6 +1,6 @@
1
1
  {
2
2
  "name": "ruvnet-kb-first",
3
- "version": "5.0.0",
3
+ "version": "6.0.0",
4
4
  "description": "RuvNet KB-First Application Builder - Build intelligent applications on expert knowledge",
5
5
  "type": "module",
6
6
  "main": "src/index.js",