Skip to main content
Glama

Kilo-Kit: Autonomous Cognitive Flow & Quality Engine for AI Coding Agents

Version: 1.9.0
Author: Kilo-Kit Team
License: Apache 2.0

Kilo-Kit is an agentic MCP runtime and curated 180-skill catalog designed to enforce grounded diagnosis, Tree-of-Thoughts architectural planning, adversarial red-teaming, 4D quality verification, and continuous SQLite self-improvement for AI coding assistants.

๐Ÿง  Core Architectural Pillars:

  1. Division of Labor (Cortex vs Limbs): Kilo-Kit acts as the high-level cognitive brain (Tree of Thoughts, 5-Whys root cause analysis, adversarial stress-testing, context compaction) while host clients handle surgical I/O.

  2. Kilo-Sentinel Supervisor & Circuit Breaker: Real-time middleware enforcing Pre-flight Grounding Locks (no editing unread files), loop tripwires (identical call and edit-thrashing detection), and SQLite trajectory logging (katl_trajectories).

  3. Triangulated Cognitive Synthesis & Low-Confidence Escalation: Combines internal SQLite memory recall, GitHub 10k+ stars patterns, and ToT DAG benchmarking (kilo_triangulate_research), with automatic subagent delegation when confidence < 0.70.

  4. Fuzzy Skill & Alias Resolver: Instant, resilient skill loading with support for aliases (brainstorming, diagnose, playwright, clean-code, tdd, grounded-research-benchmark).

  5. 4D Quality Assurance & Playwright E2E Gate: Validates Given-When-Then acceptance criteria, clean code interfaces, UI/UX aesthetics, and automated Playwright browser/DOM verification before work is marked complete.


๐Ÿ›๏ธ System Architecture & Division of Labor

Kilo-Kit enforces a strict architectural boundary between High-Level Cognitive Reasoning (Cortex) and Surgical I/O Execution (Limbs):

flowchart TD
    Clients["๐Ÿ–ฅ๏ธ Host AI Clients<br/>(Claude Code / Antigravity / Cursor / Gemini CLI)"]
    
    Sentinel["๐Ÿ›ก๏ธ Kilo-Sentinel Supervisor & Circuit Breaker<br/>(Pre-flight Grounding Lock โ€ข Loop Tripwire โ€ข Step Budget)"]
    
    subgraph Cortex["๐Ÿง  Kilo-Kit Cognitive Cortex (MCP Runtime)"]
        direction TB
        C4["๐Ÿ›๏ธ C4 5-Gate Lifecycle Controller"]
        Engines["โš™๏ธ 6 Cognitive Engines<br/>(ToT DAG โ€ข Adversarial Grill โ€ข 5-Whys โ€ข Grounded Synthesis)"]
        Skills["๐Ÿ“š 180 Curated Skills Catalog"]
        Limbs["๐Ÿ› ๏ธ Safe Execution Limbs<br/>(Atomic Write โ€ข AST Edit โ€ข Security Filtered Exec)"]
        C4 --> Engines
        C4 --> Skills
        C4 --> Limbs
    end
    
    DB[("๐Ÿ’พ SQLite Atomic Memory<br/>(cognitive_triangulations โ€ข katl_trajectories โ€ข facts)")]
    
    Clients <-->|MCP Protocol / stdio| Sentinel
    Sentinel <--> Cortex
    Cortex <--> DB

Related MCP server: CornMCP

๐Ÿ”„ C4 Cognitive Lifecycle & Low-Confidence Escalator

All agent tasks flow through a deterministic 5-Gate state machine. Unauthorized file mutations prior to Gate 3 approval are blocked at the server level:

flowchart LR
    G1["<b>Gate 1: Grounded Probe</b><br/>โ€ข Memory Recall<br/>โ€ข Codebase Probe"]
    
    G2["<b>Gate 2: Cognitive Reasoning</b><br/>โ€ข 3-Option ToT DAG<br/>โ€ข Adversarial Grill<br/>โ€ข Low-Confidence Escalator"]
    
    G3["<b>Gate 3: Approval</b><br/>โ€ข Plan Locked<br/>โ€ข Skills Injected"]
    
    G4["<b>Gate 4: Execution</b><br/>โ€ข Defense-in-Depth<br/>โ€ข Sentinel Guarded"]
    
    G5["<b>Gate 5: 4D QA</b><br/>โ€ข Playwright E2E<br/>โ€ข SQLite Reflection"]

    G1 --> G2
    G2 -->|Confidence >= 0.70| G3
    G2 -.->|Confidence < 0.70<br/>Escalation| Subagent["๐Ÿ”ฌ Research Subagent<br/>(GitHub & Docs Sandbox)"]
    Subagent -.-> G2
    G3 --> G4
    G4 --> G5
    G4 -.->|3x Loop Detected| Breaker["๐Ÿ›‘ Circuit Breaker<br/>(Supervised Reset)"]
    Breaker -.-> G1

๐Ÿ›ก๏ธ Protocol-Level Hard-Gate Enforcement

Traditional prompt rules (.cursorrules, CLAUDE.md) suffer from prompt drift. Kilo-Kit enforces safety via JSON-RPC Interceptor Middleware:

flowchart TD
    Req["๐Ÿ‘ค User Prompt"] --> Agent["๐Ÿค– Host AI Agent"]
    
    Agent -->|1. Unauthorized Edit Attempt| GateCheck{"๐Ÿ›ก๏ธ Kilo-Sentinel<br/>Pre-Flight Lock"}
    GateCheck -->|โŒ Uninitialized / Unread File| Blocked["๐Ÿ›‘ 403 Hard-Gate Blocked<br/>(Forces Planning First)"]
    
    Blocked --> Plan["๐Ÿง  Gate 1 & 2: C4 Planning<br/>(kilo_orchestrate_task + kilo_triangulate_research)"]
    Plan --> SaveDB[("๐Ÿ’พ Commit CoT to SQLite")]
    SaveDB --> Approval["๐Ÿ‘ค User Approves Plan"]
    
    Approval -->|2. Authorized Execution| GateCheck
    GateCheck -->|โœ… State = READY| Exec["๐Ÿš€ Gate 4: Safe Execution<br/>(kilo_write_file / kilo_edit_file)"]
    Exec --> Verify["โœ… Gate 5: 4D Verification & SQLite Reflection"]

๐Ÿงฐ The 24 All-in-One MCP Tools Suite

Kilo-Kit provides a complete, self-contained execution and cognitive runtime:

Category

Tool

Description

Gating & Orchestration

kilo_orchestrate_task

C4 closed-loop gate. Enforces brainstorming and cognitive steps before code mutation.

kilo_route_intent

Routes intent to best workflow chains, task modes, and rules.

kilo_get_skill

Loads curated SKILL.md workflows with token-safe truncation and session tracking.

kilo_search_skills

High-precision semantic and keyword search across 180 skills.

kilo_memory_report

Inspects persistent SQLite decisions, facts, and sessions.

kilo_remember_fact

Pins immutable operational rules and architectural decisions into SQLite memory_facts.

kilo_record_reflection

Self-Improvement: Persists reflections, correct/wrong paths, and lessons to SQLite.

kilo_route_report

Reports route telemetry, top skills, workflows, scores, and conflict penalties.

kilo_validate_skills

Validates entire skill catalog against the quality gate.

Cognitive Reasoning

kilo_triangulate_research

Grounded Synthesis & Low-Confidence Escalator: Combines SQLite memory + GitHub grounding + 3-option ToT DAG, atomically commits reasoning to SQLite, and triggers research escalation when confidence < 0.70.

kilo_think_step

Tree of Thoughts DAG: Step-by-step reasoning, 3-option trade-off matrix & hypothesis branching.

kilo_grill_plan

Adversarial Red-Teaming: Inversion, simplification, mobile touch & concurrency stress testing.

kilo_trace_root_cause

5-Whys Diagnostic Engine: Recursive causal back-propagation with regression test scaffolding.

kilo_compact_context

Cognitive Compactor: 40-70% token savings while locking invariants.

kilo_synthesize_skill

Self-Evolution: Distills solved patterns into reusable skills.

Sentinel & Supervision

kilo_sentinel_status

Supervisor Telemetry: Inspects circuit breaker state, step budget, and grounded files list.

kilo_reset_circuit_breaker

Supervised Reset: Resets tripped circuit breaker with root-cause justification.

kilo_benchmark_solution

Industry Benchmark: Audits trajectory against GitHub standards and triggers re-planning.

Safe Execution Suite

kilo_read_file

Line slicing, size capping, and repository boundary enforcement.

kilo_search_files

Glob pattern search across directory trees.

kilo_grep_code

Line-by-line regex and substring search.

kilo_write_file

Atomic write with Protocol Hard-Gate, clean-code smell audit, and secret detection.

kilo_edit_file

Targeted search-and-replace with JSON syntax & bracket balancing audit.

kilo_run_command

Defense-in-depth terminal execution with security guardrails & command injection filtering.


๐Ÿ“š 180 Curated Skills Catalog Taxonomy

Skills are organized into 6 functional modules with instant alias mapping:

skills/
โ”œโ”€โ”€ ๐Ÿ—๏ธ engineering/ (39 skills)
โ”‚   โ”œโ”€โ”€ backend-development, codebase-design, api-patterns, database-design
โ”‚   โ”œโ”€โ”€ nextjs-best-practices, react-patterns, tailwind-patterns, aspnet-core, better-auth
โ”œโ”€โ”€ ๐Ÿงฉ problem-solving/ (24 skills)
โ”‚   โ”œโ”€โ”€ sequential-thinking, root-cause-tracing, systematic-debugging
โ”‚   โ”œโ”€โ”€ collision-zone-thinking, scale-game, simplification-cascades, inversion-exercise
โ”œโ”€โ”€ ๐Ÿ“‹ productivity/ (33 skills)
โ”‚   โ”œโ”€โ”€ brainstorming, spec-driven-development, tdd-workflow, code-review
โ”‚   โ”œโ”€โ”€ grounded-research-benchmark, verification-before-completion, grill-me, subagent-driven-development
โ”œโ”€โ”€ ๐Ÿค– agent-frameworks/ (26 skills)
โ”‚   โ”œโ”€โ”€ workflow-state-machines, agent-memory, agentic-rag, multi-agent-orchestration
โ”‚   โ”œโ”€โ”€ mcp-agent-patterns, code-agent-patterns, context-optimization
โ”œโ”€โ”€ ๐Ÿ›ก๏ธ security/ (22 skills)
โ”‚   โ”œโ”€โ”€ ai-guardrails, red-team-tactics, security-best-practices, vulnerability-scanner
โ””โ”€โ”€ โ˜๏ธ devops-cloud/ (36 skills)
    โ”œโ”€โ”€ devops, server-management, chrome-devtools, performance-profiling, render-deploy

Fuzzy Alias Resolution: Calling kilo_get_skill("brainstorming") automatically loads productivity/brainstorming/SKILL.md.


โšก Quick Start & Multi-Client IDE Setup

Global Installation

npm install -g @vodailoc/kilo-kit-mcp
kilo-kit-init global

Multi-Client Configuration

1. Claude Code (~/.claude.json or project root)

{
  "mcpServers": {
    "kilo-kit": {
      "command": "kilo-kit-mcp",
      "args": []
    }
  }
}

2. Antigravity & Gemini CLI (~/.gemini/antigravity-cli/mcp_config.json)

{
  "mcpServers": {
    "kilo-kit": {
      "command": "node",
      "args": ["/usr/local/lib/node_modules/@vodailoc/kilo-kit-mcp/dist/index.js"]
    }
  }
}

3. Cursor & Windsurf (.cursor/mcp.json)

{
  "mcpServers": {
    "kilo-kit": {
      "command": "npx",
      "args": ["-y", "@vodailoc/kilo-kit-mcp"]
    }
  }
}

4. Team Repository Rollout

Generate protocol guidelines (CLAUDE.md, AGENTS.md, GEMINI.md) for your entire engineering team:

kilo-kit-init init --client all
git add CLAUDE.md AGENTS.md GEMINI.md
git commit -m "chore: configure Kilo-Kit C4 cognitive protocol"

๐Ÿ“Š Empirical Verification & Quality Benchmarks

Metric

Without Kilo-Kit (Vanilla Agent)

With Kilo-Kit v1.9.0

Verification Mechanism

Ungrounded Code Mutations

42% of sessions (modifying unread files)

0% (100% Blocked)

Server-side Pre-flight Grounding Lock

Context Window Longevity

Degrades at >30k tokens

Sustained >150k tokens

kilo_compact_context (40โ€“70% token pruning)

Silent Regression Rate

28% of PRs

< 2%

4D QA (Playwright + Given-When-Then criteria)

Tool Thrashing / Infinite Loops

Common on complex bugs

Terminated โ‰ค 3 loops

Kilo-Sentinel Loop & Thrashing Tripwire

Self-Healing & Reasoning Recall

Zero across sessions

100% SQLite Persistence

cognitive_triangulations & katl_trajectories


๐Ÿ“– Documentation Directory Index

For detailed specifications, protocol definitions, and developer guides, explore the docs/ directory:


๐Ÿงช Development, Testing & Verification

# Clone repository
git clone https://github.com/VoDaiLocz/KILO-KIT.git
cd KILO-KIT
npm install

# Run unit test suites (15/15 test files, 68/68 tests passing)
npm test

# Run full health diagnostics
npm run doctor

# Validate 100% of 180 skills against structural quality gates
node src/tools/validate-skill.js --all skills

๐Ÿ“„ License

Distributed under the Apache 2.0 License. See LICENSE for more details.

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections. Dates show when Glama detected each change.

No tool schema history has been recorded yet.

Maintenance

ActivityMaintained
ResponsivenessResponsive

Related MCP Connectors

Related MCP Servers

Latest Blog Posts

MCP directory API

We provide all the information about MCP servers via our MCP API.

curl -X GET 'https://glama.ai/api/mcp/v1/servers/VoDaiLocz/kilo-kit-mcp'

If you have feedback or need assistance with the MCP directory API, please join our Discord server