konsilio 0.3.0 → 0.3.1

package/LICENSE.md

@@ -1,24 +1,24 @@
-This is free and unencumbered software released into the public domain.
-
-Anyone is free to copy, modify, publish, use, compile, sell, or
-distribute this software, either in source code form or as a compiled
-binary, for any purpose, commercial or non-commercial, and by any
-means.
-
-In jurisdictions that recognize copyright laws, the author or authors
-of this software dedicate any and all copyright interest in the
-software to the public domain. We make this dedication for the benefit
-of the public at large and to the detriment of our heirs and
-successors. We intend this dedication to be an overt act of
-relinquishment in perpetuity of all present and future rights to this
-software under copyright law.
-
-THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
-EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
-MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
-IN NO EVENT SHALL THE AUTHORS BE LIABLE FOR ANY CLAIM, DAMAGES OR
-OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
-ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
-OTHER DEALINGS IN THE SOFTWARE.
-
+This is free and unencumbered software released into the public domain.
+
+Anyone is free to copy, modify, publish, use, compile, sell, or
+distribute this software, either in source code form or as a compiled
+binary, for any purpose, commercial or non-commercial, and by any
+means.
+
+In jurisdictions that recognize copyright laws, the author or authors
+of this software dedicate any and all copyright interest in the
+software to the public domain. We make this dedication for the benefit
+of the public at large and to the detriment of our heirs and
+successors. We intend this dedication to be an overt act of
+relinquishment in perpetuity of all present and future rights to this
+software under copyright law.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
+IN NO EVENT SHALL THE AUTHORS BE LIABLE FOR ANY CLAIM, DAMAGES OR
+OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+OTHER DEALINGS IN THE SOFTWARE.
+
 For more information, please refer to <https://unlicense.org>

package/README.md

@@ -1,170 +1,172 @@
-# Konsilio
-
-**Council of Experts** — an MCP server that runs draft plans through a multi-persona architectural review before you write a single line of code.
-
-## TL;DR
-
-Konsilio lets you run a draft plan through a panel of AI experts (security, performance, DevOps, etc.) before writing code.
-
-It returns a structured blueprint so your coding model can implement it cleanly in one pass.
-
-## The Problem
-
-You're mid-task and realize you need a plan: *"Why is this code not working? We need more logs → logging should be added → let me write a proposal for logging."*
-
-You could feed that plan directly into a coding model and hope it gets it right. Or you could catch blind spots early by having specialists review it first.
-
-## The Solution
-
-Konsilio implements a **multi-stage consulting pipeline**:
-
-1. **Write a proposal** — a short document describing what you want to build or change
-2. **Feed it to the Council** — multiple expert personas (security, performance, DevOps, testing, etc.) review the plan in parallel, each from their own angle
-3. **4-phase consolidation** — a lead agent extracts claims, critiques contradictions, makes accept/reject decisions, and synthesizes a final blueprint
-4. **Implement with a clean slate** — hand the blueprint to a coding-focused model. It starts with empty context and executes the plan in one pass
-
-**Rinse. Repeat. Rewind Selectah!**
-
-## Usage
-
-```bash
-# Install
-npm install -g konsilio
-
-# Configure (edit konsilio.json with your personas and models)
-# Set OPENROUTER_API_KEY in your environment
-
-# Use via MCP in your AI coding tool
-# Tool: consult_council
-# Params: draft_plan, tech_stack (optional), context_constraints (optional)
-```
-
-### Example
-
-```json
-{
-  "draft_plan": "Add structured logging to all API endpoints with correlation IDs",
-  "tech_stack": "Node.js, Express, Pino, PostgreSQL",
-  "context_constraints": "Must run on Proxmox LXC, no external dependencies"
-}
-```
-
-Returns a blueprint with architecture directives, edge cases, constraints, and numbered next steps.
-
-## Configuration
-
-```json
-{
-  "personas": {
-    "enabled": ["security", "performance", "devops", "test-architect"]
-  },
-  "models": {
-    "experts": "google/gemini-2.5-flash-lite",
-    "lead": "google/gemini-2.5-pro"
-  }
-}
-```
-
-Default personas: `security`, `performance`, `ux-dx`, `devops`, `typescript`, `graph-dba`, `node-fullstack`, `dev-tooling`, `distributed-systems`, `test-architect`
-
-```mermaid
-flowchart TD
-    A[Identify Need During Development] --> B[Write Proposal Document]
-    B --> C[Submit to Konsilio Council]
-    
-    subgraph Council["Council Review (Parallel)"]
-        C --> E1[🔒 Security Expert]
-        C --> E2[⚡ Performance Expert]
-        C --> E3[🚀 DevOps Expert]
-        C --> E4[🧪 Test Architect]
-        C --> E5[🎨 UX/DX Expert]
-        C --> E6[🛠️ Dev Tooling Expert]
-    end
-    
-    E1 --> F[Prose Reports]
-    E2 --> F
-    E3 --> F
-    E4 --> F
-    E5 --> F
-    E6 --> F
-    
-    F --> G[Formatter: prose → structured JSON]
-    
-    subgraph Consolidation["4-Phase Consolidation"]
-        G --> H1[📋 Extraction Lead<br/>Extract all claims]
-        H1 --> H2[🔍 Critique Lead<br/>Find contradictions]
-        H2 --> H3[⚖️ Decision Lead<br/>Accept/Reject findings]
-        H3 --> H4[📝 Synthesis Lead<br/>Assemble blueprint]
-    end
-    
-    H4 --> I[Final Blueprint<br/>with edge cases, constraints, next steps]
-    I --> J[Hand to Coding Model<br/>Clean slate, empty context]
-    J --> K[Implement in One Pass]
-    
-    style Council fill:#f0f4ff,stroke:#4a6cf7,stroke-width:2px
-    style Consolidation fill:#fff8f0,stroke:#f7a44a,stroke-width:2px
-    style I fill:#e8f5e9,stroke:#4caf50,stroke-width:2px
-```
-
-## Architecture
-
-### Expert Personas
-
-Each expert is a focused agent with a specific lens:
-
-| Persona | Focus |
-|---------|-------|
-| 🔒 Security | Auth, injection, rate limiting, data exposure |
-| ⚡ Performance | Bottlenecks, caching, query patterns, scaling |
-| 🚀 DevOps | Deployment, observability, infrastructure |
-| 🧪 Test Architect | Coverage, edge cases, testability |
-| 🎨 UX/DX | API design, developer experience, ergonomics |
-| 🛠️ Dev Tooling | Build pipeline, linting, CI/CD |
-
-Experts output **prose** — free-form analysis without JSON constraints.
-**Note:** User-defined experts can be created by adding an `expert.md` file to `/data/personas/`
-
-### The Lead
-
-The lead agent runs a 4-phase consolidation pipeline:
-
-1. **Extraction** — pulls all claims from expert reports into a unified list
-2. **Critique** — identifies contradictions, unsupported claims, and gaps
-3. **Decision** — explicitly accepts or rejects each finding
-4. **Synthesis** — assembles accepted findings into a coherent blueprint
-
-### The Formatter
-
-A dedicated `gpt-4o-mini` instance converts expert prose to structured JSON using OpenAI's `response_format` feature. This keeps experts focused on analysis, not syntax.
-
-## Why This Works
-
-### 1. Specialized Experts Outperform Generalists
-
-A council of focused expert agents, each primed with a specific role, surfaces more issues at lower cost than a single frontier model:
-
-- **Chan et al. (2023)** — *"ChatEval: Towards Better LLM-based Evaluators through Multi-Agent Debate"* — showed that multiple role-conditioned agents debating produces more accurate evaluations than a single powerful model. [arXiv:2308.07201](https://arxiv.org/abs/2308.07201)
-
-- **Du et al. (2023)** — *"Improving Factuality and Reasoning in Language Models through Multiagent Debate"* — demonstrated that multi-agent debate significantly improves accuracy over single-model inference, with agents catching each other's errors. [arXiv:2305.14325](https://arxiv.org/abs/2305.14325)
-
-- **Liang et al. (2023)** — *"Ensemble Learning for Large Language Models"* — found that combining multiple specialized models outperforms a single large model on complex tasks while using fewer total tokens. [arXiv:2305.07881](https://arxiv.org/abs/2305.07881)
-
-### 2. Separation of Analysis and Formatting
-
-When a model is asked to return structured data directly, its reasoning quality degrades. Konsilio solves this by having experts write **prose first**, then using a dedicated formatter:
-
-- **Dhuliawala et al. (2023)** — *"Chain-of-Verification Reduces Hallucination in Large Language Models"* — demonstrates that separating generation from verification significantly reduces errors. This is the core principle behind Konsilio's two-stage design. [arXiv:2309.11495](https://arxiv.org/abs/2309.11495)
-
-- **Willison et al. (2023)** — *"Jsonformer: A Bulletproof Way to Generate Structured JSON from Language Models"* — documents how JSON schema constraints during generation interfere with reasoning quality, recommending a two-stage approach: generate content freely, then format. [arXiv:2305.15087](https://arxiv.org/abs/2305.15087)
-
-- **Su et al. (2024)** — *"The Impact of Reasoning Step Length on Large Language Models"* — shows that adding structural constraints (like JSON output requirements) during reasoning reduces the model's ability to explore solution spaces, increasing error rates. [arXiv:2401.04991](https://arxiv.org/abs/2401.04991)
-
-This is why Konsilio uses `gpt-4o-mini` as a **dedicated formatter** — it converts prose to JSON without the expert models having to juggle both reasoning and formatting simultaneously.
-
-### 3. Clean-Slate Implementation
-
-By separating planning from execution, the coding model starts with **zero context overhead**. It receives a complete blueprint and executes — no need to re-derive the plan or carry forward the reasoning history. This reduces token costs and avoids context-window pollution.
-
-# License
-
-This software is released under the Unlicense: do whatever you want with it.
+# Konsilio
+
+[![npm version](https://img.shields.io/npm/v/konsilio.svg)](https://www.npmjs.com/package/konsilio)
+[![npm downloads](https://img.shields.io/npm/dm/konsilio.svg)](https://www.npmjs.com/package/konsilio)
+[![license](https://img.shields.io/npm/l/konsilio.svg)](LICENSE.md)
+
+Konsilio lets you run a draft plan through a panel of AI experts (security, performance, DevOps, etc.) before writing code.
+
+It returns a structured blueprint so your coding model can implement it cleanly in one pass.
+
+## The Problem
+
+You're mid-task and realize you need a plan:
+
+*"Why is this code not working? We need more logs → logging should be added → let me write a proposal for logging."*
+
+You could feed that plan directly into a coding model and hope it gets it right. Or you could catch blind spots early by having specialists review it first.
+
+## The Solution
+
+Konsilio implements a **multi-stage consulting pipeline**:
+
+1. **Write a proposal** — a short document describing what you want to build or change
+2. **Feed it to the Council** — multiple expert personas (security, performance, DevOps, testing, etc.) review the plan in parallel, each from their own angle
+3. **4-phase consolidation** — a lead agent extracts claims, critiques contradictions, makes accept/reject decisions, and synthesizes a final blueprint
+4. **Implement with a clean slate** — hand the blueprint to a coding-focused model. It starts with empty context and executes the plan in one pass
+
+**Rinse. Repeat. Rewind Selectah!**
+
+## Usage
+
+```bash
+# Install
+npm install -g konsilio
+
+# Configure (edit konsilio.json with your personas and models)
+# Set OPENROUTER_API_KEY in your environment
+
+# Use via MCP in your AI coding tool
+# Tool: consult_council
+# Params: draft_plan, tech_stack (optional), context_constraints (optional)
+```
+
+### Example
+
+```json
+{
+  "draft_plan": "Add structured logging to all API endpoints with correlation IDs",
+  "tech_stack": "Node.js, Express, Pino, PostgreSQL",
+  "context_constraints": "Must run on Proxmox LXC, no external dependencies"
+}
+```
+
+Returns a blueprint with architecture directives, edge cases, constraints, and numbered next steps.
+
+## Configuration
+
+```json
+{
+  "personas": {
+    "enabled": ["security", "performance", "devops", "test-architect"]
+  },
+  "models": {
+    "experts": "google/gemini-2.5-flash-lite",
+    "lead": "google/gemini-2.5-pro"
+  }
+}
+```
+
+Default personas: `security`, `performance`, `ux-dx`, `devops`, `typescript`, `graph-dba`, `node-fullstack`, `dev-tooling`, `distributed-systems`, `test-architect`
+
+```mermaid
+flowchart TD
+    A[Identify Need During Development] --> B[Write Proposal Document]
+    B --> C[Submit to Konsilio Council]
+    
+    subgraph Council["Council Review (Parallel)"]
+        C --> E1[🔒 Security Expert]
+        C --> E2[⚡ Performance Expert]
+        C --> E3[🚀 DevOps Expert]
+        C --> E4[🧪 Test Architect]
+        C --> E5[🎨 UX/DX Expert]
+        C --> E6[🛠️ Dev Tooling Expert]
+    end
+    
+    E1 --> F[Prose Reports]
+    E2 --> F
+    E3 --> F
+    E4 --> F
+    E5 --> F
+    E6 --> F
+    
+    F --> G[Formatter: prose → structured JSON]
+    
+    subgraph Consolidation["4-Phase Consolidation"]
+        G --> H1[📋 Extraction Lead<br/>Extract all claims]
+        H1 --> H2[🔍 Critique Lead<br/>Find contradictions]
+        H2 --> H3[⚖️ Decision Lead<br/>Accept/Reject findings]
+        H3 --> H4[📝 Synthesis Lead<br/>Assemble blueprint]
+    end
+    
+    H4 --> I[Final Blueprint<br/>with edge cases, constraints, next steps]
+    I --> J[Hand to Coding Model<br/>Clean slate, empty context]
+    J --> K[Implement in One Pass]
+    
+    style Council fill:#f0f4ff,stroke:#4a6cf7,stroke-width:2px
+    style Consolidation fill:#fff8f0,stroke:#f7a44a,stroke-width:2px
+    style I fill:#e8f5e9,stroke:#4caf50,stroke-width:2px
+```
+
+## Architecture
+
+### Expert Personas
+
+Each expert is a focused agent with a specific lens:
+
+| Persona | Focus |
+|---------|-------|
+| 🔒 Security | Auth, injection, rate limiting, data exposure |
+| ⚡ Performance | Bottlenecks, caching, query patterns, scaling |
+| 🚀 DevOps | Deployment, observability, infrastructure |
+| 🧪 Test Architect | Coverage, edge cases, testability |
+| 🎨 UX/DX | API design, developer experience, ergonomics |
+| 🛠️ Dev Tooling | Build pipeline, linting, CI/CD |
+
+Experts output **prose** — free-form analysis without JSON constraints.
+**Note:** User-defined experts can be created by adding an `expert.md` file to `/data/personas/`
+
+### The Lead
+
+The lead agent runs a 4-phase consolidation pipeline:
+
+1. **Extraction** — pulls all claims from expert reports into a unified list
+2. **Critique** — identifies contradictions, unsupported claims, and gaps
+3. **Decision** — explicitly accepts or rejects each finding
+4. **Synthesis** — assembles accepted findings into a coherent blueprint
+
+### The Formatter
+
+A dedicated `gpt-4o-mini` instance converts expert prose to structured JSON using OpenAI's `response_format` feature. This keeps experts focused on analysis, not syntax.
+
+## Why This Works
+
+### 1. Specialized Experts Outperform Generalists
+
+A council of focused expert agents, each primed with a specific role, surfaces more issues at lower cost than a single frontier model:
+
+- **Chan et al. (2023)** — *"ChatEval: Towards Better LLM-based Evaluators through Multi-Agent Debate"* — showed that multiple role-conditioned agents debating produces more accurate evaluations than a single powerful model. [arXiv:2308.07201](https://arxiv.org/abs/2308.07201)
+
+- **Du et al. (2023)** — *"Improving Factuality and Reasoning in Language Models through Multiagent Debate"* — demonstrated that multi-agent debate significantly improves accuracy over single-model inference, with agents catching each other's errors. [arXiv:2305.14325](https://arxiv.org/abs/2305.14325)
+
+- **Liang et al. (2023)** — *"Ensemble Learning for Large Language Models"* — found that combining multiple specialized models outperforms a single large model on complex tasks while using fewer total tokens. [arXiv:2305.07881](https://arxiv.org/abs/2305.07881)
+
+### 2. Separation of Analysis and Formatting
+
+When a model is asked to return structured data directly, its reasoning quality degrades. Konsilio solves this by having experts write **prose first**, then using a dedicated formatter:
+
+- **Dhuliawala et al. (2023)** — *"Chain-of-Verification Reduces Hallucination in Large Language Models"* — demonstrates that separating generation from verification significantly reduces errors. This is the core principle behind Konsilio's two-stage design. [arXiv:2309.11495](https://arxiv.org/abs/2309.11495)
+
+- **Willison et al. (2023)** — *"Jsonformer: A Bulletproof Way to Generate Structured JSON from Language Models"* — documents how JSON schema constraints during generation interfere with reasoning quality, recommending a two-stage approach: generate content freely, then format. [arXiv:2305.15087](https://arxiv.org/abs/2305.15087)
+
+- **Su et al. (2024)** — *"The Impact of Reasoning Step Length on Large Language Models"* — shows that adding structural constraints (like JSON output requirements) during reasoning reduces the model's ability to explore solution spaces, increasing error rates. [arXiv:2401.04991](https://arxiv.org/abs/2401.04991)
+
+This is why Konsilio uses `gpt-4o-mini` as a **dedicated formatter** — it converts prose to JSON without the expert models having to juggle both reasoning and formatting simultaneously.
+
+### 3. Clean-Slate Implementation
+
+By separating planning from execution, the coding model starts with **zero context overhead**. It receives a complete blueprint and executes — no need to re-derive the plan or carry forward the reasoning history. This reduces token costs and avoids context-window pollution.
+
+# License
+
+This software is released under the Unlicense: do whatever you want with it.

package/build/prompts/personas/test-quoted.md

@@ -1,14 +1,14 @@
----
-id: "test-quoted"
-name: "Test Quoted Persona"
-emoji: "🧪"
-focusAreas:
-  - "Testing quoted values"
-  - "YAML parsing"
-domains:
-  - "test"
----
-
-## Anti-Patterns
-
+---
+id: "test-quoted"
+name: "Test Quoted Persona"
+emoji: "🧪"
+focusAreas:
+  - "Testing quoted values"
+  - "YAML parsing"
+domains:
+  - "test"
+---
+
+## Anti-Patterns
+
 - "This is a quoted anti-pattern"

package/build/services/formatter.service.js

@@ -84,24 +84,24 @@ export class FormatterService {
      * Build the system prompt for the formatter
      */
     buildFormatterSystemPrompt(personaId) {
-        return `You are a precise JSON formatter. Your job is to convert expert analysis prose into a structured JSON output.
-
-Extract the following from the prose:
-- findings: Specific issues with severity, component, issue description, and mitigation steps
-- risks: Potential risks with category, probability, impact, and description
-- missingAssumptions: Assumptions not stated in the analysis
-- dependencies: External dependencies or prerequisites mentioned
-
-Rules:
-- Use severity values: CRITICAL, HIGH, MEDIUM, LOW
-- Use category values: security, performance, operational, ux, technical-debt
-- Use probability/impact values: high, medium, low
-- Each finding must have a unique id (use kebab-case like "auth-rate-limit")
-- Each risk must have a unique id
-- Be precise and complete - capture all findings from the prose
-- Do not add findings not supported by the prose
-- Set personaId to: ${personaId}
-
+        return `You are a precise JSON formatter. Your job is to convert expert analysis prose into a structured JSON output.
+
+Extract the following from the prose:
+- findings: Specific issues with severity, component, issue description, and mitigation steps
+- risks: Potential risks with category, probability, impact, and description
+- missingAssumptions: Assumptions not stated in the analysis
+- dependencies: External dependencies or prerequisites mentioned
+
+Rules:
+- Use severity values: CRITICAL, HIGH, MEDIUM, LOW
+- Use category values: security, performance, operational, ux, technical-debt
+- Use probability/impact values: high, medium, low
+- Each finding must have a unique id (use kebab-case like "auth-rate-limit")
+- Each risk must have a unique id
+- Be precise and complete - capture all findings from the prose
+- Do not add findings not supported by the prose
+- Set personaId to: ${personaId}
+
 Output ONLY valid JSON. No markdown, no explanations.`;
     }
     /**

package/package.json

@@ -1,68 +1,68 @@
-{
-  "name": "konsilio",
-  "version": "0.3.0",
-  "description": "Konsilio MCP server runs draft plans through a multi-persona architectural review.",
-  "keywords": [
-    "mcp",
-    "model-context-protocol",
-    "code-review",
-    "architecture",
-    "ai",
-    "multi-agent",
-    "expert-system",
-    "planning",
-    "consulting",
-    "blueprint"
-  ],
-  "homepage": "https://github.com/edzillion/konsilio#readme",
-  "bugs": {
-    "url": "https://github.com/edzillion/konsilio/issues"
-  },
-  "repository": {
-    "type": "git",
-    "url": "git+https://github.com/edzillion/konsilio.git"
-  },
-  "license": "Unlicense",
-  "author": "edzillion",
-  "type": "module",
-  "exports": {
-    ".": "./build/index.js",
-    "./schema": "./konsilio.schema.json"
-  },
-  "main": "build/index.js",
-  "types": "./build/index.d.ts",
-  "bin": {
-    "konsilio": "build/index.js"
-  },
-  "files": [
-    "build/",
-    "konsilio.schema.json"
-  ],
-  "scripts": {
-    "build": "tsc -p tsconfig.build.json && node scripts/copy-assets.mjs",
-    "dev": "tsc --watch",
-    "start": "node build/index.js",
-    "start:server": "node build/server.js",
-    "inspect": "npx @modelcontextprotocol/inspector node /g/dev/konsilio/build/index.js",
-    "typecheck": "tsc --noEmit",
-    "history": "node build/cli/history.js",
-    "test": "vitest run --reporter=verbose",
-    "test-agent": "vitest run --reporter=json | jq \".testResults[] | {name, status, errors}\"",
-    "test:watch": "vitest",
-    "test:coverage": "vitest run --coverage"
-  },
-  "dependencies": {
-    "@modelcontextprotocol/sdk": "^1.28.0",
-    "pino": "^10.3.1",
-    "sql.js": "^1.14.1",
-    "zod": "^4.3.6"
-  },
-  "devDependencies": {
-    "@modelcontextprotocol/inspector": "^0.21.1",
-    "@types/node": "^25.5.0",
-    "@types/sql.js": "^1.4.11",
-    "@vitest/coverage-v8": "^4.1.2",
-    "typescript": "^6.0.2",
-    "vitest": "^4.1.2"
-  }
-}
+{
+  "name": "konsilio",
+  "version": "0.3.1",
+  "description": "Konsilio MCP server runs draft plans through a multi-persona architectural review.",
+  "keywords": [
+    "mcp",
+    "model-context-protocol",
+    "code-review",
+    "architecture",
+    "ai",
+    "multi-agent",
+    "expert-system",
+    "planning",
+    "consulting",
+    "blueprint"
+  ],
+  "homepage": "https://github.com/edzillion/konsilio#readme",
+  "bugs": {
+    "url": "https://github.com/edzillion/konsilio/issues"
+  },
+  "repository": {
+    "type": "git",
+    "url": "git+https://github.com/edzillion/konsilio.git"
+  },
+  "license": "Unlicense",
+  "author": "edzillion",
+  "type": "module",
+  "exports": {
+    ".": "./build/index.js",
+    "./schema": "./konsilio.schema.json"
+  },
+  "main": "build/index.js",
+  "types": "./build/index.d.ts",
+  "bin": {
+    "konsilio": "build/index.js"
+  },
+  "files": [
+    "build/",
+    "konsilio.schema.json"
+  ],
+  "scripts": {
+    "build": "tsc -p tsconfig.build.json && node scripts/copy-assets.mjs",
+    "dev": "tsc --watch",
+    "start": "node build/index.js",
+    "start:server": "node build/server.js",
+    "inspect": "npx @modelcontextprotocol/inspector node /g/dev/konsilio/build/index.js",
+    "typecheck": "tsc --noEmit",
+    "history": "node build/cli/history.js",
+    "test": "vitest run --reporter=verbose",
+    "test-agent": "vitest run --reporter=json | jq \".testResults[] | {name, status, errors}\"",
+    "test:watch": "vitest",
+    "test:coverage": "vitest run --coverage"
+  },
+  "dependencies": {
+    "@modelcontextprotocol/sdk": "^1.28.0",
+    "pino": "^10.3.1",
+    "sql.js": "^1.14.1",
+    "zod": "^4.3.6"
+  },
+  "devDependencies": {
+    "@modelcontextprotocol/inspector": "^0.21.1",
+    "@types/node": "^25.5.0",
+    "@types/sql.js": "^1.4.11",
+    "@vitest/coverage-v8": "^4.1.2",
+    "typescript": "^6.0.2",
+    "vitest": "^4.1.2"
+  }
+}