kilo-kit-mcp

Imported from VoDaiLocz/kilo-kit-mcp (AGENTS.md). Install upstream with npx skills add VoDaiLocz/kilo-kit-mcp. Copyright stays with the author.

Kilo-Kit C4 v3.0 Cognitive Protocol

Applies to: Antigravity, Gemini CLI, Claude Code, OpenAI Codex

🧠 DIVISION OF LABOR (VỎ NÃO & CHÂN TAY):

• Kilo-Kit MCP (Cortex): High-level cognitive reasoning, 5-Whys root cause tracing, Tree-of-Thoughts DAG planning, adversarial red-teaming, context compaction, skill delivery, and SQLite learning memory.
• Execution Limbs: Perform surgical changes with defense-in-depth using your native tools or Kilo-Kit tools with Kilo-Sentinel supervision.

🏛️ MANDATORY 5-GATE COGNITIVE PROTOCOL:

1. Gate 1 (Hard-Gate & Grounded Probe):

   • Call MCP tool kilo_orchestrate_task with the user's request as your FIRST action.
   • Perform read-only exploratory probing (view_file, grep_search, or kilo_read_file) to build a Grounded Diagnostic Model.

2. Gate 2 (Substantive Cognitive Reasoning & Benchmarking):

   • For Features / Architecture: MUST call kilo_think_step (DAG with 3 trade-off options) and kilo_grill_plan (Adversarial Red-Team stress test). Call kilo_benchmark_solution to audit against industry standards.
   • For Bugs / Errors: MUST call kilo_trace_root_cause with { errorLog, failingFile } to extract the 5-Whys causal chain before writing any fix.
   • For Context Overload (>4 files read): Call kilo_compact_context to lock invariants and prevent attention degradation.

3. Gate 3 (Approval & Skill Delivery):

   • Call kilo_orchestrate_task with brainstormingApproved=true.
   • Load required workflow skills using kilo_get_skill (supports aliases like brainstorming, diagnose, playwright, tdd, clean-code).
   • (Optional) Check supervisor status via kilo_sentinel_status.

4. Gate 4 (Surgical Implementation & Defense-in-Depth):

   • Apply problem-solving/defense-in-depth (validation at input, logic, and persistence layers).
   • Maintain engineering/clean-code standards with zero bloat.

5. Gate 5 (4D Quality Assurance & Reflection):

   • Dimension 1 (Spec Fidelity): Verify against user requirements using Given-When-Then acceptance criteria (productivity/spec-driven-development).
   • Dimension 2 (Clean Code): Verify deep module interfaces and zero dead code (engineering/clean-code).
   • Dimension 3 (UX & Aesthetics): Verify responsive layout and mobile touch safety (design/aesthetic, design/mobile-design).
   • Dimension 4 (Empirical & Playwright): Run compile/build commands AND execute tests before claiming completion.
   • Continuous Reflection: Call kilo_record_reflection with sessionId to persist lessons learned into SQLite memory.

📢 MANDATORY INTER-TOOL NARRATION (ZERO SILENT CHAINS):

• Iron Law: NEVER execute consecutive tool calls without printing a 1-2 sentence intermediate update directly to the user. The terminal user CANNOT inspect internal MCP return payloads.
• Micro-Narration Protocol before issuing next tool call:
  • [DECISION]: State what you just concluded, selected, or verified cụ thể.
  • [NEXT]: State what tool you are calling next and for what purpose.