v1.9.0: Grounded Cognitive Triangulation, Reasoning SQLite Store & Low-Confidence Escalation

🚀 Kilo-Kit v1.9.0 Release Notes

🌟 Major Features:

1. New Tool: kilo_triangulate_research (24th MCP Tool):
   • Implements Triangulated Cognitive Synthesis combining Internal SQLite Memory, External GitHub Grounding, 3-Option ToT DAG Trade-Offs, and Low-Confidence Research Escalation.
2. Atomic Reasoning Persistence in SQLite:
   • Created cognitive_triangulations table in SQLite to persistently record all reasoning, DAG options, and chosen architectures before code modification.
3. Low-Confidence Subagent Escalation:
   • Automatically detects when confidenceScore < 0.70 or requiresResearchEscalation=true, emitting high-priority escalation directives to delegate research into an isolated sandbox without context pollution.
4. New Core Skill: skills/grounded-research-benchmark/SKILL.md:
   • Standard operating procedure for Triangulated Grounding, DAG evaluation, and self-evolving knowledge.
5. Full System Health:
   • 8/8 Doctor checks passing, 15/15 test suites passing (68/68 unit tests, 25/25 live tool audit assertions).

v1.8.9: Universal Trajectory Ledger for All 23 Tools

🚀 Kilo-Kit v1.8.9 Release Notes

• Universal Trajectory Logging: Connected all 23 MCP tools directly to Sentinel's katl_trajectories ledger.
• Full Traceability: Every tool call across orchestrator, facts, reflections, skills, router, cognitive DAGs, and execution limbs is now permanently tracked in SQLite.
• Fixed Trajectory Fields: Corrected skill ID reference in kilo_synthesize_skill and failure streak metric in kilo_sentinel_status.

v1.8.8: Remember Fact Tool, SQLite WAL Mode & Enhanced Decision Engine

🚀 Kilo-Kit v1.8.8 Release Notes

🌟 Key Improvements:

1. New Tool: kilo_remember_fact:
   • Enables explicit cross-session persistence of project architecture rules, testing configurations, and operating defaults directly into SQLite memory_facts.
2. SQLite Performance & Composite Indexing:
   • Activated PRAGMA journal_mode = WAL and PRAGMA synchronous = NORMAL for 5-10x faster concurrent read/write throughput.
   • Added composite indexes on learning_reflections(task_mode, created_at), orchestration_sessions(state, created_at), workflow_outcomes(session_id, created_at), and memory_facts(kind).
3. Fixed Memory Confirmation Filtering:
   • Fixed bug where non-blocking advisory reflection suggestions (requiresConfirmation: false) were incorrectly treated as pending confirmations.
4. Enriched Memory Report (kilo_memory_report):
   • Formatter now renders the complete 5-pillar SQLite overview including Top 10 Learning Reflections with lessons learned and avoided paths.
5. Persistent Route Analytics:
   • Route decisions now default to persistent recording in ~/.kilo-kit/decision-trail.jsonl.
6. 100% Comprehensive Audit:
   • All 23 tools verified passing in live C4 5-Gate lifecycle.

v1.8.7: Autonomous Learning Memory Recall & Cognitive Trajectory Tracing

🧠 Kilo-Kit v1.8.7 Release Notes

🌟 Key Improvements:

1. Closed-Loop Autonomous Learning Recall:
   • Upgraded memory suggestion engine in orchestration-memory.ts to automatically query past learning_reflections from SQLite based on task mode and recency.
   • Eliminates the write-only memory loophole: Past lessons learned are now injected directly into Gate 1 (kilo_orchestrate_task) under ## Memory Suggestions.
2. Complete Cognitive Trajectory Tracing:
   • Sentinel now records trajectory steps in katl_trajectories for cognitive tools: kilo_think_step, kilo_grill_plan, kilo_trace_root_cause, kilo_compact_context, kilo_benchmark_solution.
   • Full 360-degree observability of agent reasoning cycles, durations, and risk verdicts in SQLite.
3. Empirical Verification:
   • 100% test coverage passed (15 test suites, 68 unit tests, 8/8 Doctor checks).

v1.8.6: Mandatory Inter-Tool Narration & Decision Visibility

🚀 Highlights in v1.8.6

• Zero Silent Chains (Iron Law): Prohibits AI agents from executing consecutive tool calls without intermediate user-facing text updates on the terminal.
• Inter-Tool Narration Directives: Automatically appends [MANDATORY INTER-TOOL NARRATION DIRECTIVE] with [DECISION] and [NEXT] requirements to all Gate 1, Gate 2, Gate 3, Gate 5 tool outputs.
• Recency Attention Steering: Leverages LLM recency bias to ensure models immediately explain their choices to users before triggering the next tool.
• Synchronized Protocols: Updated AGENTS.md, GEMINI.md, and CLAUDE.md with standard [DECISION] & [NEXT] micro-narration protocols.
• 100% Verified: 68/68 unit tests passed and 10/10 end-to-end narration assertions verified.

v1.8.5: Code-Enforced Observability & Terminal Call Labels

🚀 Highlights in v1.8.5

• Code-Enforced Tool Call Observability: Added label parameter (Zod schema validation, min 10 chars) across Gate 2 (kilo_think_step, kilo_grill_plan, kilo_trace_root_cause) and Gate 5 (kilo_record_reflection).
• Prominent Header Echo: MCP responses automatically echo > 🏷️ **[label]** prominently at the top of the output, making tool reasoning transparent in the terminal UI.
• Nudge Warning Tip: When agents fail to provide a descriptive label, the tool responds with > ⚠️ Tip: thêm \label` param để user thấy bạn đang làm gì`.
• Full 5-Gate Telemetry: Passed all 68 unit tests and live 8/8 E2E verification assertions.

Kilo-Kit MCP v1.8.4 - Full 5-Gate Cognitive Protocol & Supervisory Control Layer

🚀 Kilo-Kit MCP v1.8.4 Release Notes

🧠 Full 5-Gate Cognitive Protocol & Supervisory Control Layer

• Cortex vs Limbs Architecture: Clean separation where Kilo-Kit acts as the high-level Cognitive Cortex (3-option Trade-Off DAG planning, 5-Whys root cause tracing, adversarial grilling, context compaction, and skill delivery) while host agents use native limbs or Kilo-Kit tools with Kilo-Sentinel supervision.
• Supervisory & Control Layer (, , , ): Fully integrated into state machine transitions and nextAction prompts across all session workflows.
• Context Overload Compactor: Automatically guides agents to lock architectural invariants via kilo_compact_context when exploring deep codebases (>4 files).
• Closed-Loop GitHub Industry Benchmarking: Audits architectural plans against top open-source industry patterns via kilo_benchmark_solution before code modification.
• Sentinel Anomaly & Loop Protection: Real-time loop detection, edit thrashing prevention, pre-flight grounding lock, and SQLite trajectory logging.
• Self-Improvement Memory: Persists correct approaches, wrong paths avoided, and lessons learned into SQLite (kilo_record_reflection).

🛠️ Deep Code Audit & 15+ Latent Bug Fixes

• Fixed stateful regex lastIndex in grep tool for 100% match rate.
• Fixed substring fallback in glob file search.
• Increased max file read limit from 32KB to 512KB.
• Fixed analytics score adjustment in skill routing.
• Fixed SQLite file descriptor cleanup in Doctor diagnostics.
• Synced in-memory sort orders and structured clone support in orchestration memory.

🧪 Verification & Quality

• 15/15 test suites passed, 68/68 tests passed (100%).
• Kilo-Kit Doctor: 8/8 checks passed (HEALTHY).

Kilo-Kit MCP v1.8.3 - Force Kilo-Kit Execution Limbs

Kilo-Kit MCP v1.8.3

• Fix: Updated C4 Protocol in AGENTS.md and init.ts to explicitly forbid the use of native Edit/Bash limbs.
• Architecture: AI Agents must now route all file editing and terminal executions through kilo_edit_file, kilo_write_file, and kilo_run_command. This ensures that Kilo-Sentinel Circuit Breaker and Grounding Lock can intercept and supervise all executions.

Kilo-Kit MCP v1.8.2 - Industry Standard Team Rollout

Kilo-Kit MCP v1.8.2

• New Feature: Added kilo-kit-init global for 1-click global deployment across AI IDEs.
• Documentation: Restructured README to follow Configuration-as-Code industry standard.
• Fixes: Cleaned up legacy workspace bootstrapping docs.

Kilo-Kit MCP v1.6.0 - All-in-One 18 MCP Tools Suite & Universal 1-Click Setup

🚀 What's New in Kilo-Kit v1.6.0

⚡ 1. Universal 1-Click Auto-Installer

• Run npx -y @vodailoc/kilo-kit-mcp setup to automatically detect and register Kilo-Kit across Antigravity CLI, Cursor IDE, Windsurf, Claude Code, and Claude Desktop in 1 second!

---

🛡️ 2. Protocol-Level Server-Side Hard-Gate (Zero-File Dependency)

• Enforces strict C4 workflow gating directly at the MCP JSON-RPC protocol level.
• Any unauthorized file modification (kilo_write_file, kilo_edit_file) or terminal command execution (kilo_run_command) without an approved sessionId in state: ready is instantly blocked by the MCP server.
• Eliminates prompt drift and prevents agents from skipping planning.

---

🧰 3. Complete 18 All-in-One MCP Tools Suite

1. Gating & Orchestration: kilo_orchestrate_task, kilo_route_intent, kilo_memory_report, kilo_get_skill, kilo_search_skills, kilo_route_report, kilo_validate_skills.
2. Safe Execution Suite: kilo_read_file, kilo_search_files, kilo_grep_code, kilo_write_file, kilo_edit_file, kilo_run_command.
3. 5 Cognitive Reasoning Engines:
   • 🧠 kilo_think_step: Tree of Thoughts DAG reasoning with hypothesis branching and revisions.
   • 🥊 kilo_grill_plan: Automated adversarial red-teaming (Inversion, Simplification Cascades, Blast Radius).
   • 🔍 kilo_trace_root_cause: 5-Whys recursive causal backward-propagation.
   • 🗜️ kilo_compact_context: Token economy compactor (40-70% reduction) with invariant locking.
   • 🧬 kilo_synthesize_skill: Autonomous self-evolving skill distillation into standardized SKILL.md.

---

🧪 4. Comprehensive Testing & Quality Gate

• Passed all 12 test suites (51/51 unit tests 100% green).
• Validated all 177/177 Skills against the quality gate.
• Clean Stdio smoke test passed.

---

📦 Quick Start

# 1-Click setup for all AI clients
npx -y @vodailoc/kilo-kit-mcp setup

# Or add to Claude Code
claude mcp add kilo-kit -- npx -y @vodailoc/kilo-kit-mcp