@techwavedev/agi-agent-kit 1.6.3 → 1.6.5

package/CHANGELOG.md

@@ -4,6 +4,28 @@ All notable changes to this project will be documented in this file.
 
 The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/),
 and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
+## [1.6.5] - 2026-03-19
+
+### Added
+
+- **Skill Self-Improvement (Karpathy Loop)** — Autonomous test → improve → commit/reset cycle for continuous skill quality improvement. Inspired by Andrej Karpathy's "auto-research" concept:
+  - `execution/run_skill_eval.py` — Binary assertion runner with 18 assertion types (`contains`, `regex_match`, `max_words`, `has_yaml_frontmatter`, `no_trailing_whitespace`, etc.)
+  - `execution/karpathy_loop.py` — Autonomous loop orchestrator with git commit/reset integration, dry-run mode, and status reporting
+  - `eval/evals.json` standard — Structured binary assertion format for objective skill quality measurement
+- **Skill Creator: Step 8** — Added "Self-Improvement Loop (Karpathy Loop)" step to `SKILL_skillcreator.md` with methodology, assertion examples, and auto-generation guidance
+- **Eval Directory Scaffolding** — `init_skill.py` now creates `eval/evals.json` with 4 starter assertions (frontmatter, description quality, structure, no-placeholders) for every new skill
+- **Example evals** — `skills/qdrant-memory/eval/evals.json` added as reference implementation (17/17 passing)
+
+### Documentation
+
+- **`AGENTS.md`** — Added "Skill Self-Improvement (Karpathy Loop)" section with quick start commands, assertion types reference, and key rules. Updated directory structure to show `eval/`. Added eval scripts to Key Scripts section.
+
+## [1.6.4] - 2026-03-14
+
+### Fixed
+
+- **CI publish pipeline** — `release_gate.py` no longer hangs on interactive prompts when running in GitHub Actions. All `input()` calls are now CI-aware: auto-fail with a clear error message in non-interactive environments.
+- **CHANGELOG version check** — release gate now exits non-zero with an actionable message instead of prompting when the version is missing from CHANGELOG.
 
 ## [1.6.3] - 2026-03-14
 

package/README.md

@@ -16,6 +16,8 @@
 [![Antigravity IDE](https://img.shields.io/badge/Antigravity-DeepMind-red)](https://github.com/techwavedev/agi-agent-kit)
 [![AdaL CLI](https://img.shields.io/badge/AdaL%20CLI-SylphAI-pink)](https://sylph.ai/)
 [![OpenClaw](https://img.shields.io/badge/OpenClaw-CLI-teal)](https://github.com/openclaw/openclaw)
+[![Kiro](https://img.shields.io/badge/Kiro-AWS-yellow)](https://kiro.dev)
+[![MCP Compatible](https://img.shields.io/badge/MCP-Compatible-blueviolet)](https://modelcontextprotocol.io)
 [![Buy me a coffee](https://img.shields.io/badge/Buy%20me%20a-coffee-d13610?logo=buymeacoffee&logoColor=white)](https://www.buymeacoffee.com/eltonmachado)
 
 **AGI Agent Kit** is the enterprise-grade scaffolding that turns any AI coding assistant into a **deterministic production machine**. While LLMs are probabilistic (90% accuracy per step = 59% over 5 steps), this framework forces them through a **3-Layer Architecture** — Intent → Orchestration → Execution — where business logic lives in tested scripts, not hallucinated code.
@@ -28,7 +30,8 @@ Most AI coding setups give you a prompt and hope for the best. AGI Agent Kit giv
 - 🎯 **19 Specialist Agents** — Domain-bounded experts (Frontend, Backend, Security, Mobile, Game Dev...) with enforced file ownership
 - ⚡ **1,191 Curated Skills** — 4 core + 89 professional + 1,098 community skills across 16 domain categories
 - 🔒 **Verification Gates** — No task completes without evidence. TDD enforcement. Two-stage code review.
-- 🌐 **9 Platforms, One Config** — Write once, run on Claude Code, Gemini CLI, Codex CLI, Cursor, Copilot, OpenCode, AdaL CLI, Antigravity IDE, OpenClaw
+- 🌐 **10 Platforms, One Config** — Write once, run on Claude Code, Gemini CLI, Codex CLI, Cursor, Copilot, OpenCode, AdaL CLI, Antigravity IDE, OpenClaw, Kiro
+- 🔌 **MCP Compatible** — Exposes memory + cross-agent coordination as MCP tools for Claude Desktop and any chat-interface client
 
 ```bash
 npx @techwavedev/agi-agent-kit init
@@ -85,12 +88,14 @@ python3 skills/plugin-discovery/scripts/platform_setup.py --project-dir .
 | **Distributed Agent Auth**    | HMAC-SHA256 signing, hash anchoring, project access control via shared Qdrant (Hyperledger Aries optional) |
 | **Real-Time Agent Events**    | Apache Pulsar event bus for push notifications between agents — graceful degradation if unavailable |
 | **Hybrid Memory**             | Qdrant vectors + BM25 keywords with weighted score merge (95% token savings)                  |
-| **Platform-Adaptive**         | Auto-detects Claude Code, Gemini CLI, Codex CLI, Cursor, Copilot, OpenCode, AdaL, Antigravity |
+| **Platform-Adaptive**         | Auto-detects Claude Code, Gemini CLI, Codex CLI, Cursor, Copilot, OpenCode, AdaL, Antigravity, Kiro |
+| **MCP Compatible**            | Memory + cross-agent coordination exposed as MCP tools (`execution/mcp_server.py`) for Claude Desktop and any MCP client |
 | **Multi-Agent Orchestration** | Agent Teams, subagents, Powers, or sequential personas — adapts to platform                   |
 | **Structured Plan Execution** | Batch or subagent-driven execution with two-stage review (spec + quality)                     |
 | **TDD Enforcement**           | Iron-law RED-GREEN-REFACTOR cycle — no production code without failing test                   |
 | **Verification Gates**        | Evidence before claims — no completion without fresh verification output                      |
 | **Self-Healing Workflows**    | Agents read error logs, patch scripts, and update directives automatically                    |
+| **Skill Self-Improvement**    | Karpathy Loop: autonomous test → improve → commit/reset cycle with 18 binary assertion types  |
 | **One-Shot Setup**            | Platform detection + project stack scan + auto-configuration in one command                   |
 
 ---
@@ -106,7 +111,7 @@ The agi framework adopts all best patterns from [obra/superpowers](https://githu
 | Systematic Debugging         |        ✅        | ✅ Adapted + `debugger` agent  |
 | Verification Gates           |        ✅        | ✅ Adapted + 12 audit scripts  |
 | Two-Stage Code Review        |        ✅        |  ✅ Adapted into orchestrator  |
-| Multi-Platform Orchestration |  ❌ Claude only  |         ✅ 9 platforms         |
+| Multi-Platform Orchestration |  ❌ Claude only  |         ✅ 10 platforms        |
 | Semantic Memory (Qdrant)     |        ❌        |    ✅ 90-100% token savings    |
 | 19 Specialist Agents         |        ❌        |      ✅ Domain boundaries      |
 | Agent Boundary Enforcement   |        ❌        |     ✅ File-type ownership     |
@@ -114,7 +119,8 @@ The agi framework adopts all best patterns from [obra/superpowers](https://githu
 | Memory-First Protocol        |        ❌        |       ✅ Auto cache-hit        |
 | Skill Creator + Catalog      |        ❌        |  ✅ 1,191 composable skills   |
 | Platform Setup Wizard        |        ❌        |       ✅ One-shot config       |
-| Multi-Platform Symlinks      |  ❌ Claude only  |         ✅ 9 platforms         |
+| Multi-Platform Symlinks      |  ❌ Claude only  |        ✅ 10 platforms         |
+| MCP Server                   |        ❌        |   ✅ Memory + coordination     |
 
 ---
 
@@ -205,6 +211,7 @@ Skills are installed to the canonical `skills/` directory and symlinked to each
 | **GitHub Copilot**  | N/A (paste)       | `COPILOT.md`     | Manual paste into context           |
 | **OpenCode**        | `.agent/skills/`  | `OPENCODE.md`    | Sequential personas via `@agent`    |
 | **AdaL CLI**        | `.adal/skills/`   | `AGENTS.md`      | Auto-load on demand                 |
+| **Kiro (AWS)**      | `.kiro/skills/`   | `.kiro/steering/agents.md` | Full agentic orchestration   |
 
 Run `/setup` to auto-detect and configure your platform, or use the setup script directly:
 
@@ -599,9 +606,11 @@ If these community skills help you, consider [starring the original repo](https:
 | **Blockchain Agent Trust & Tenancy** | ✅ Shipped   | HMAC-SHA256 signed writes, hash anchoring, project access control, audit trail — all via shared Qdrant `agent_auth` collection. Optional W3C DID via Hyperledger Aries ACA-Py 1.5.0. 36/36 tests. ([docs](./docs/blockchain-auth.md)) |
 | **Event-Driven Agent Streaming**    | ✅ Shipped    | Apache Pulsar event bus with auto-publish on `memory_manager.py store`. Project-scoped topics, graceful degradation. 19/19 tests. ([docs](./docs/agent-events.md))                                                        |
 | **Memory Mode Tiers**               | ✅ Shipped    | Solo → Team → Pro progression. Backward-compatible upgrades, no data migration. BM25 auto-synced from shared Qdrant on boot. ([docs](./docs/memory-modes.md))                                                            |
+| **MCP Compatibility**               | ✅ Shipped    | Memory + cross-agent coordination exposed as MCP tools via `execution/mcp_server.py` (13 tools) and `skills/qdrant-memory/mcp_server.py` (6 tools). Pure chat clients (Claude Desktop) get full memory access. ([docs](./docs/mcp-compatibility.md)) |
+| **Platform-Adaptive Orchestration** | ✅ Shipped    | 10 platforms share one `AGENTS.md` via symlinks (Claude Code, Gemini CLI, Codex CLI, Cursor, Copilot, OpenCode, AdaL, Antigravity, OpenClaw, Kiro). Each uses its native orchestration strategy automatically.           |
+| **Workflow Engine**                 | ✅ Shipped    | `execution/workflow_engine.py` executes `data/workflows.json` playbooks as guided multi-skill sequences with progress tracking, skip/abort, and state persistence in `.tmp/playbook_state.json`.                         |
+| **Skill Self-Improvement**          | ✅ Shipped    | Karpathy Loop: `run_skill_eval.py` (18 binary assertion types) + `karpathy_loop.py` (autonomous test/improve/commit/reset). Skills include `eval/evals.json` for objective quality measurement.                           |
 | **Control Tower Orchestrator**      | 🚧 Active    | Basic dispatcher for agent registration and heartbeat via Qdrant (`control_tower.py`). Needs dedicated docs, test coverage, and integration with `session_boot`.                                                          |
-| **Workflow Engine**                 | 📋 Planned   | Execute `data/workflows.json` playbooks as guided multi-skill sequences with progress tracking. Data schema exists, execution script not yet implemented.                                                                 |
-| **Platform-Adaptive Orchestration** | 🔬 Design    | Use platform-native orchestration when available (Anthropic Agent Teams, Gemini sub-agents) and fall back to Qdrant-based Control Tower dispatch otherwise. Unified interface across providers.                            |
 | **Secrets Management (Vault)**      | 🔬 Design    | HashiCorp Vault integration for secure secret sharing. Agents authenticate via Ed25519 keypair, access tenant-scoped secrets. Zero long-lived credentials.                                                                |
 
 ---

package/README.pt-BR.md

@@ -16,6 +16,8 @@
 [![Antigravity IDE](https://img.shields.io/badge/Antigravity-DeepMind-red)](https://github.com/techwavedev/agi-agent-kit)
 [![AdaL CLI](https://img.shields.io/badge/AdaL%20CLI-SylphAI-pink)](https://sylph.ai/)
 [![OpenClaw](https://img.shields.io/badge/OpenClaw-CLI-teal)](https://github.com/openclaw/openclaw)
+[![Kiro](https://img.shields.io/badge/Kiro-AWS-yellow)](https://kiro.dev)
+[![MCP Compatible](https://img.shields.io/badge/MCP-Compatible-blueviolet)](https://modelcontextprotocol.io)
 [![Buy me a coffee](https://img.shields.io/badge/Buy%20me%20a-coffee-d13610?logo=buymeacoffee&logoColor=white)](https://www.buymeacoffee.com/eltonmachado)
 
 **AGI Agent Kit** é o scaffolding de nível empresarial que transforma qualquer assistente de código com IA em uma **máquina de produção determinística**. Enquanto LLMs são probabilísticos (90% de precisão por etapa = 59% ao longo de 5 etapas), este framework os força através de uma **Arquitetura de 3 Camadas** — Intenção → Orquestração → Execução — onde a lógica de negócio vive em scripts testados, não em código alucinado.
@@ -28,7 +30,8 @@ A maioria das configurações de IA para código te dá um prompt e torce pelo m
 - 🎯 **19 Agentes Especialistas** — Experts delimitados por domínio (Frontend, Backend, Segurança, Mobile, Game Dev...) com propriedade de arquivos forçada
 - ⚡ **1.191 Skills Curadas** — 4 essenciais + 89 profissionais + 1.098 comunitárias em 16 categorias de domínio
 - 🔒 **Portões de Verificação** — Nenhuma tarefa é concluída sem evidência. Enforcement de TDD. Revisão de código em duas etapas.
-- 🌐 **9 Plataformas, Uma Configuração** — Escreva uma vez, execute no Claude Code, Gemini CLI, Codex CLI, Cursor, Copilot, OpenCode, AdaL CLI, Antigravity IDE, OpenClaw
+- 🌐 **10 Plataformas, Uma Configuração** — Escreva uma vez, execute no Claude Code, Gemini CLI, Codex CLI, Cursor, Copilot, OpenCode, AdaL CLI, Antigravity IDE, OpenClaw, Kiro
+- 🔌 **Compatível com MCP** — Expõe memória + coordenação entre agentes como ferramentas MCP para Claude Desktop e qualquer cliente de interface de chat
 
 ```bash
 npx @techwavedev/agi-agent-kit init
@@ -84,7 +87,8 @@ python3 skills/plugin-discovery/scripts/platform_setup.py --project-dir .
 | **Execução de Plano Estruturado** | Execução em lote ou orientada por subagentes com revisão em duas etapas (spec + qualidade)    |
 | **Enforcement de TDD**            | Ciclo RED-GREEN-REFACTOR inquebrantável — sem código de produção sem teste falhando           |
 | **Portões de Verificação**        | Evidência antes de afirmações — sem conclusão sem saída de verificação atualizada             |
-| **Adaptativo à Plataforma**       | Auto-detecta Claude Code, Gemini CLI, Codex CLI, Cursor, Copilot, OpenCode, AdaL, Antigravity |
+| **Adaptativo à Plataforma**       | Auto-detecta Claude Code, Gemini CLI, Codex CLI, Cursor, Copilot, OpenCode, AdaL, Antigravity, Kiro |
+| **Compatível com MCP**            | Memória + coordenação entre agentes expostas como ferramentas MCP (`execution/mcp_server.py`) para Claude Desktop e qualquer cliente MCP |
 | **Orquestração Multi-Agente**     | Agent Teams, subagentes, Powers, ou personas sequenciais — adapta-se à plataforma             |
 | **Memória Híbrida**               | Vetores Qdrant + palavras-chave BM25 com mesclagem de pontuação ponderada (95% de economia)   |
 | **Workflows Auto-Reparáveis**     | Agentes leem logs de erro, corrigem scripts e atualizam diretivas automaticamente             |
@@ -103,7 +107,7 @@ O framework AGI adota todos os melhores padrões do [obra/superpowers](https://g
 | Debugging Sistemático            |        ✅        |    ✅ Adaptado + agente `debugger`    |
 | Portões de Verificação           |        ✅        |   ✅ Adaptado + 12 scripts de audit   |
 | Revisão de Código em Duas Etapas |        ✅        |      ✅ Adaptado no orquestrador      |
-| Orquestração Multi-Plataforma    | ❌ Apenas Claude |           ✅ 9 plataformas            |
+| Orquestração Multi-Plataforma    | ❌ Apenas Claude |          ✅ 10 plataformas            |
 | Memória Semântica (Qdrant)       |        ❌        |     ✅ 90-100% economia de tokens     |
 | 19 Agentes Especialistas         |        ❌        |       ✅ Fronteiras de domínio        |
 | Reforço de Fronteiras de Agente  |        ❌        |  ✅ Propriedade por tipo de arquivo   |
@@ -111,7 +115,8 @@ O framework AGI adota todos os melhores padrões do [obra/superpowers](https://g
 | Protocolo Memory-First           |        ❌        |           ✅ Auto cache-hit           |
 | Criador de Skills + Catálogo     |        ❌        |       ✅ 1.191 skills combináveis       |
 | Assistente de Configuração       |        ❌        |     ✅ Configuração em um comando     |
-| Symlinks Multi-Plataforma        | ❌ Apenas Claude |           ✅ 9 plataformas            |
+| Symlinks Multi-Plataforma        | ❌ Apenas Claude |          ✅ 10 plataformas            |
+| Servidor MCP                     |        ❌        |    ✅ Memória + coordenação           |
 
 ---
 
@@ -202,6 +207,7 @@ Skills são instaladas no diretório canônico `skills/` e vinculadas via symlin
 | **GitHub Copilot**  | N/A (colar)        | `COPILOT.md`         | Colar manualmente no contexto        |
 | **OpenCode**        | `.agent/skills/`   | `OPENCODE.md`        | Personas sequenciais via `@agent`    |
 | **AdaL CLI**        | `.adal/skills/`    | `AGENTS.md`          | Carregamento automático sob demanda  |
+| **Kiro (AWS)**      | `.kiro/skills/`    | `.kiro/steering/agents.md` | Orquestração agêntica completa |
 
 Execute `/setup` para auto-detectar e configurar sua plataforma, ou use o script de configuração diretamente:
 
@@ -506,12 +512,17 @@ Se essas skills comunitárias te ajudam, considere [dar uma estrela no repositó
 
 ## 🗺️ Roadmap
 
-| Funcionalidade                         | Status       | Descrição                                                                                                                                                                                                                                                           |
-| -------------------------------------- | ------------ | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
-| **Memória Federada de Agentes**        | 🔬 Design    | Compartilhamento de conhecimento entre agentes via coleções Qdrant com escopo de projeto. Agentes no mesmo projeto leem decisões, erros e padrões uns dos outros — construindo inteligência coletiva entre sessões e plataformas.                                   |
-| **Memória Autenticada por Blockchain** | 🔬 Design    | Camada de confiança criptográfica para memória compartilhada usando blockchains empresariais (Hyperledger Fabric, MultiChain ou Quorum) — auto-hospedado, sem taxas, sem criptomoeda. Escritas de agentes são assinadas, hashes de conteúdo são ancorados on-chain. |
-| **Streaming de Agentes Event-Driven**  | 🔬 Design    | Comunicação em tempo real entre agentes via Kafka/Flink. Agentes publicam decisões e observações em tópicos, habilitando workflows reativos — ex: agente de segurança aciona remediação quando o agente de varredura publica achados.                               |
-| **Motor de Workflows**                 | 📋 Planejado | Execução de playbooks `data/workflows.json` como sequências multi-skill guiadas com rastreamento de progresso e lógica de ramificação.                                                                                                                              |
+| Funcionalidade                             | Status       | Descrição                                                                                                                                                                                                                                                           |
+| ------------------------------------------ | ------------ | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
+| **Memória Federada de Agentes**            | ✅ Entregue  | Compartilhamento de conhecimento entre agentes via Qdrant compartilhado. Multi-tenancy com isolamento por desenvolvedor, flag `--shared` para visibilidade em equipe. 15/15 testes. ([docs](./docs/memory-modes.md)) |
+| **Autenticação Blockchain de Agentes**     | ✅ Entregue  | Escritas assinadas com HMAC-SHA256, ancoragem de hash, controle de acesso por projeto, trilha de auditoria — tudo via coleção Qdrant `agent_auth`. DID W3C opcional via Hyperledger Aries ACA-Py 1.5.0. 36/36 testes. ([docs](./docs/blockchain-auth.md)) |
+| **Streaming Event-Driven de Agentes**      | ✅ Entregue  | Barramento de eventos Apache Pulsar com auto-publicação no `memory_manager.py store`. Tópicos com escopo de projeto, degradação graciosa. 19/19 testes. ([docs](./docs/agent-events.md)) |
+| **Níveis de Modo de Memória**              | ✅ Entregue  | Progressão Solo → Team → Pro. Upgrades com compatibilidade retroativa, sem migração de dados. BM25 auto-sincronizado do Qdrant compartilhado no boot. ([docs](./docs/memory-modes.md)) |
+| **Compatibilidade MCP**                    | ✅ Entregue  | Memória + coordenação entre agentes expostas como ferramentas MCP via `execution/mcp_server.py` (13 ferramentas) e `skills/qdrant-memory/mcp_server.py` (6 ferramentas). ([docs](./docs/mcp-compatibility.md)) |
+| **Orquestração Adaptativa à Plataforma**   | ✅ Entregue  | 10 plataformas compartilham um `AGENTS.md` via symlinks (Claude Code, Gemini CLI, Codex CLI, Cursor, Copilot, OpenCode, AdaL, Antigravity, OpenClaw, Kiro). Cada uma usa sua estratégia de orquestração nativa automaticamente. |
+| **Motor de Workflows**                     | ✅ Entregue  | `execution/workflow_engine.py` executa playbooks `data/workflows.json` como sequências multi-skill guiadas com rastreamento de progresso, pular/abortar e persistência de estado em `.tmp/playbook_state.json`. |
+| **Control Tower Orchestrator**             | 🚧 Ativo     | Dispatcher básico para registro de agentes e heartbeat via Qdrant (`control_tower.py`). Precisa de docs dedicados e cobertura de testes. |
+| **Gerenciamento de Segredos (Vault)**      | 🔬 Design    | Integração HashiCorp Vault para compartilhamento seguro de segredos. Agentes autenticam via keypair Ed25519, acessam segredos com escopo de tenant. Zero credenciais de longa duração. |
 
 ---
 

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@techwavedev/agi-agent-kit",
-  "version": "1.6.3",
+  "version": "1.6.5",
   "description": "Enterprise-Grade Agentic Framework - Modular skill-based AI assistant toolkit with deterministic execution, semantic memory, and platform-adaptive orchestration.",
   "bin": {
     "agi-agent-kit": "./bin/init.js"

package/templates/base/AGENTS.md

@@ -1,6 +1,8 @@
 # Agent Instructions
 
-> `CLAUDE.md`, `GEMINI.md`, `OPENCODE.md`, `COPILOT.md`, and `OPENCLAW.md` are symlinks to this file, so the same instructions load in any AI environment.
+> `CLAUDE.md`, `GEMINI.md`, `OPENCODE.md`, `COPILOT.md`, and `OPENCLAW.md` are symlinks to this file.
+> Kiro reads `.kiro/steering/agents.md` (also a symlink here).
+> The same instructions load in any AI coding agent environment.
 
 ---
 
@@ -213,6 +215,17 @@ Memory types: `decision`, `code`, `error`, `technical`, `conversation`
 - ✅ You discovered a technical insight or API quirk
 - ✅ You completed a multi-step workflow
 
+#### Handling Conversation Compaction (MANDATORY)
+
+When you compact the chat or clear the context window to save tokens (such as when running `/compact` or automatically dropping early history), that context is permanently lost unless you explicitly save it. You MUST generate a summary of the vital context and push it to Qdrant immediately before the compaction:
+
+```bash
+python3 execution/memory_manager.py store \
+  --content "Pre-compaction summary: <insert key context, decisions, and current state>" \
+  --type conversation \
+  --project <project-name>
+```
+
 #### After Completing a Complex Task
 
 ```bash
@@ -753,6 +766,17 @@ python3 execution/cross_agent_context.py broadcast --agent "<name>" --message "<
 python3 execution/cross_agent_context.py pending --agent "<name>" --project agi-agent-kit
 ```
 
+### MCP Servers
+
+Two MCP servers expose the framework to Claude Desktop, Antigravity, Cursor, Copilot, and any MCP client:
+
+| Server | File | Scope |
+|---|---|---|
+| `agi-framework` | `execution/mcp_server.py` | Memory + cross-agent + health (13 tools) |
+| `qdrant-memory` | `skills/qdrant-memory/mcp_server.py` | Direct Qdrant skill ops (6 tools) |
+
+> See `docs/mcp-compatibility.md` for full setup, config examples, and compatibility matrix.
+
 ### Rules
 
 See `.agent/rules/framework_dev_rules.md` — enforces root-is-source-of-truth, mandatory sync, private file protection.

package/templates/base/CHANGELOG.md

@@ -5,16 +5,71 @@ All notable changes to this project will be documented in this file.
 The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/),
 and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
-## [1.5.4] - 2026-02-22
+## [1.6.3] - 2026-03-14
 
 ### Added
 
-- **Agent Teams & Dynamic State Handoff** — Implemented an advanced, purely deterministic sub-agent orchestration framework (closes #20):
-  - **`dispatch_agent_team.py`** — New orchestrator dynamically chains multiple targeted sub-agents, handling state transfer without agent loops or hallucinations.
-  - **Dynamic State Handoff (Pattern 6)** — Agents can pass complex intermediate data schemas directly through Qdrant semantic memory, which is pulled into the context of the next agent automatically.
-  - **Orchestrator Enforcement** — Orchestrator reads `handoff_state` explicitly requiring `next_steps` and `validation_requirements`, validating that sub-agents correctly divide tasks and dictate downstream validation.
-  - **Comprehensive Test Harness** — `execution/run_test_scenario.py` runs 6 sequential and parallel test scenarios, evaluating orchestrator logic, cross-domain parallel runs, and failure-recovery paths.
-  - **Full Team Patterns** — Built `documentation_team` (Writer, Reviewer, Changelog. Updater), `code_review_team` (Spec, Quality), `qa_team`, and `build_deploy_team`.
+- **MCP Compatibility Layer** — The framework is now consumable as MCP servers by Claude Desktop, Antigravity, Cursor, Copilot, OpenCode, OpenClaw, and any MCP-compatible client. Two servers provided:
+  - `execution/mcp_server.py` (`agi-framework`) — 13 tools covering the full execution layer: memory auto-query, store, retrieve, cache, list, cross-agent coordination (store/sync/status/handoff/broadcast/pending), and session health check.
+  - `skills/qdrant-memory/mcp_server.py` (`qdrant-memory`) — 6 tools wrapping the skill's Python modules directly (no subprocess, no external package).
+- **MCP Skill Scaffolding** — `init_skill.py` now auto-generates `mcp_server.py` + `mcp_tools.json` for every new skill. Template lives at `skill-creator/templates/mcp_server_template.py`.
+
+- **Kiro (AWS) compatibility** — Added `.kiro/steering/agents.md` symlink → `AGENTS.md`. Kiro now loads the same agent instructions as Claude Code, Antigravity, and all other coding agents.
+
+### Design Decision
+
+MCP support is **purely additive** — no existing files modified, no validator changes. Rationale: `quick_validate.py` enforces an allowlist `{name, description, license, allowed-tools, metadata}`; adding new top-level SKILL.md keys would break all existing skills. MCP lives in separate files (`mcp_server.py`, `mcp_tools.json`) alongside the skill structure. The `metadata:` field (already allowed) is available for lightweight MCP hints.
+
+**MCP scope clarification:** All coding agents with bash access (Antigravity, Claude Code, Kiro, Cursor, OpenCode, Copilot) were already fully compatible via the symlinked `AGENTS.md`. Agent teams, sub-agents, skills, and memory all work natively for these agents. MCP is only needed for pure chat interfaces (Claude Desktop, etc.) with no bash execution capability.
+
+### Documentation
+
+- **`docs/mcp-compatibility.md`** — New. Full MCP reference: tool catalog, client compatibility matrix, architecture decision rationale, setup for all agents.
+- **`skills/qdrant-memory/SKILL.md`** — Updated MCP Quick Start section to reference `mcp_server.py` instead of the external `@qdrant/mcp-server-qdrant` package.
+- **`AGENTS.md`** — Added MCP Servers table under Framework Self-Development → Key Scripts.
+- **`docs/execution/mcp_server.md`** — New. Documents all 13 tools, transport protocol, environment variables.
+- **`skills/SKILLS_CATALOG.md`** — Regenerated; `qdrant-memory` now shows `MCP Server: mcp_server.py` row.
+
+## [1.6.2] - 2026-03-14
+
+### Added
+
+- **Contextual AGI Integration Blocks (1,200 Extended Skills)** — All 1,200 extended skills in `templates/skills/extended/` now include contextually adapted AGI Framework Integration blocks. Each block is domain-specific and provides: Memory-First Protocol (skill-specific Qdrant queries), Storing Results (concrete real-world examples), Multi-Agent Collaboration (skill-specific cross-agent sharing), and a 4th domain-specific section (e.g., Token Versioning for design-system, Brand Source of Truth for brand, Signed Audit Trail for security, Playbook Integration for devops, Asset Registry for banner design).
+
+### Fixed
+
+- **Python 3.11 f-string compatibility** — Resolved `SyntaxError` in `templates/skills/extended/frontend/ui-ux-pro-max/scripts/design_system.py`: backslash inside f-string expression extracted to intermediate variable.
+
+## [1.6.1] - 2026-03-14
+
+### Added
+
+- **Distributed Agent Authentication** — Replaced local SQLite auth storage with shared Qdrant `agent_auth` collection. Identities, access grants, content hashes, and audit trails are now distributed across all agents sharing a Qdrant instance. 8 payload indexes for efficient filtering. Deterministic UUIDs for idempotent operations. (36/36 tests)
+- **Hyperledger Aries Integration** — Replaced deprecated MultiChain with Hyperledger Aries ACA-Py 1.5.0 (OpenWallet Foundation). W3C DID identity, Ed25519 signing, official Docker image (`ghcr.io/openwallet-foundation/acapy-agent:1.5.0`). Optional add-on — HMAC-SHA256 works without Aries.
+- **Memory Mode Tiers** — Three operational modes (`MEMORY_MODE`): Solo (single user), Team (multi-tenancy with developer isolation), Pro (blockchain auth + access control). All backward-compatible, no data migration needed.
+- **Real-Time Agent Events** — Apache Pulsar event bus for push notifications between agents. Auto-publishes on `memory_manager.py store` (team/pro modes). 9 event types. Project-scoped topics (`persistent://agi/memory/<project>`). Graceful degradation when Pulsar unavailable. (19/19 tests)
+- **BM25 Auto-Sync on Boot** — `session_boot.py` now automatically syncs the local BM25 keyword index from shared Qdrant data, ensuring every machine has consistent hybrid search results.
+- **Docker Compose for Pulsar** — `docker-compose.pulsar.yml` with `apachepulsar/pulsar:4.1.3`, lightweight standalone mode (256-512MB heap), health checks.
+- **Docker Compose for Aries** — `docker-compose.aries.yml` with official ACA-Py 1.5.0 image, wallet configuration.
+
+### Changed
+
+- **`blockchain_auth.py`** — Complete rewrite: `QdrantAuthStore` replaces SQLite, `AriesClient` replaces MultiChain. All auth data stored in shared Qdrant collection.
+- **`session_boot.py`** — Now reports Pulsar status, Aries status, and BM25 sync results. Auto-syncs BM25 from Qdrant on every boot.
+- **`memory_manager.py`** — Auto-publishes Pulsar events on store (team/pro). Health check includes events status.
+- **`session_init.py`** — Creates `agent_auth` Qdrant collection with 4-dimensional vectors and 8 payload indexes.
+
+### Documentation
+
+- **`docs/memory-modes.md`** — Complete rewrite. Why each mode, when to use, scenario comparisons, data storage table. Every feature verified with test evidence.
+- **`docs/blockchain-auth.md`** — Complete rewrite. Technology comparison (MultiChain vs Ethereum vs Aries), Qdrant vs SQLite rationale, trust model diagram, all CLI commands verified.
+- **`docs/agent-events.md`** — Complete rewrite. Pulsar vs Redis/RabbitMQ/Kafka comparison, verified capabilities table, corrected dependency requirements.
+
+### Removed
+
+- **MultiChain** — Deprecated (last commit 2023, no maintained Docker image). Replaced by Hyperledger Aries.
+- **SQLite auth storage** — Replaced by shared Qdrant `agent_auth` collection for distributed consistency.
+
 ## [1.6.0] - 2026-03-12
 
 ### Added

package/templates/base/execution/dispatch_agent_team.py

@@ -27,7 +27,6 @@ import json
 import sys
 import uuid
 import os
-import subprocess
 from pathlib import Path
 from datetime import datetime, timezone
 
@@ -103,29 +102,12 @@ def load_subagent_directive(root: Path, subagent_id: str) -> dict:
 
 
 def build_manifest(team_id: str, payload: dict, subagents: list, root: Path) -> dict:
-    """Build the full dispatch manifest with identity tagging."""
+    """Build the full dispatch manifest."""
     run_id = str(uuid.uuid4())[:8]
-    
-    # Resolve developer identity
-    developer_id = os.environ.get("AGI_DEVELOPER_ID")
-    if not developer_id:
-        try:
-            result = subprocess.run(
-                ["git", "config", "user.email"],
-                capture_output=True, text=True, timeout=5
-            )
-            if result.returncode == 0 and result.stdout.strip():
-                developer_id = result.stdout.strip()
-        except (FileNotFoundError, subprocess.TimeoutExpired):
-            pass
-    if not developer_id:
-        developer_id = f"{os.environ.get('USER', 'unknown')}@{os.environ.get('HOSTNAME', 'local')}"
-
     return {
         "team": team_id,
         "run_id": run_id,
         "dispatched_at": datetime.now(timezone.utc).isoformat(),
-        "developer_id": developer_id,
         "payload": payload,
         "sub_agents": [load_subagent_directive(root, sa) for sa in subagents],
         "execution_mode": "sequential",
@@ -134,8 +116,8 @@ def build_manifest(team_id: str, payload: dict, subagents: list, root: Path) ->
             f"Pass the original payload JSON as context. If a sub-agent works on a divided task, it MUST "
             f"return a 'handoff_state' object containing its state, 'next_steps' for the following agent, "
             f"and 'validation_requirements' for what must be tested. YOU MUST actively verify this plan, "
-            f"store the state as raw JSON to Qdrant memory via `python3 execution/memory_manager.py store "
-            f"--shared` tagged with '{run_id}' against conflicts, AND pass it directly to the next sequential "
+            f"store the state as raw JSON to Qdrant memory via `python3 execution/memory_manager.py store` "
+            f"tagged with '{run_id}' against conflicts, AND pass it directly to the next sequential "
             f"or testing sub-agent so they execute the required validation and next steps precisely. "
             f"Store final results to memory with tag '{team_id}'."
         ),
@@ -143,7 +125,7 @@ def build_manifest(team_id: str, payload: dict, subagents: list, root: Path) ->
         "memory_store": (
             f"python3 execution/memory_manager.py store "
             f"--content \"{team_id} dispatched run {run_id}\" "
-            f"--type decision --tags {team_id} agent-team --shared"
+            f"--type decision --tags {team_id} agent-team"
         )
     }
 
@@ -166,8 +148,7 @@ def store_to_memory(root: Path, manifest: dict, dry_run: bool) -> bool:
             sys.executable, str(memory_script), "store",
             "--content", f"Agent team dispatched: {manifest['team']} run {manifest['run_id']}",
             "--type", "decision",
-            "--tags", manifest["team"], "agent-team-dispatch",
-            "--shared"
+            "--tags", manifest["team"], "agent-team-dispatch"
         ],
         capture_output=True,
         text=True,

package/templates/base/execution/enforce_teams.py

@@ -0,0 +1,124 @@
+#!/usr/bin/env python3
+"""
+enforce_teams.py — Session-level mandatory team checklist enforcer.
+
+Called by Claude Code hooks at key moments to remind about mandatory
+agent team runs. Tracks state in .tmp/session_teams_run.json.
+
+Usage:
+  python3 execution/enforce_teams.py check-commit   # after git commit
+  python3 execution/enforce_teams.py mark <team>    # mark a team as run
+  python3 execution/enforce_teams.py session-end    # Stop hook — final audit
+  python3 execution/enforce_teams.py reset          # new session
+"""
+
+import json
+import sys
+import os
+from datetime import datetime
+from pathlib import Path
+
+STATE_FILE = Path(".tmp/session_teams_run.json")
+
+MANDATORY_AFTER_CODE_CHANGE = [
+    "documentation_team",
+    "code_review_team",
+]
+
+MANDATORY_AFTER_COMMIT = [
+    "documentation_team",
+]
+
+MANDATORY_PER_SESSION = [
+    "cross_agent_broadcast",
+    "sync_to_template_check",
+    "memory_usage_proof",
+]
+
+
+def load_state():
+    STATE_FILE.parent.mkdir(parents=True, exist_ok=True)
+    if STATE_FILE.exists():
+        return json.loads(STATE_FILE.read_text())
+    return {"teams_run": [], "session_start": datetime.now().isoformat(), "commits_since_review": 0}
+
+
+def save_state(state):
+    STATE_FILE.write_text(json.dumps(state, indent=2))
+
+
+def mark(team: str):
+    state = load_state()
+    if team not in state["teams_run"]:
+        state["teams_run"].append(team)
+    if team == "code_review_team":
+        state["commits_since_review"] = 0
+    save_state(state)
+    print(f"✅ Marked as run: {team}")
+
+
+def check_commit():
+    """Called after git commit. Prints mandatory reminders."""
+    state = load_state()
+    state["commits_since_review"] = state.get("commits_since_review", 0) + 1
+    save_state(state)
+
+    run = state["teams_run"]
+    missing = [t for t in MANDATORY_AFTER_CODE_CHANGE if t not in run]
+
+    if missing:
+        print("\n⚠️  MANDATORY TEAMS NOT YET RUN THIS SESSION:")
+        for t in missing:
+            cmd = f"python3 execution/dispatch_agent_team.py --team {t} --payload '{{...}}'"
+            print(f"   ❌ {t}  →  {cmd}")
+        print("\nDo NOT proceed to merge/publish until these are complete.")
+        print("Mark complete with: python3 execution/enforce_teams.py mark <team>\n")
+    else:
+        print("✅ All mandatory teams run. Safe to proceed.")
+
+
+def session_end():
+    """Called at Stop. Final audit."""
+    state = load_state()
+    run = state["teams_run"]
+
+    all_mandatory = list(set(MANDATORY_AFTER_CODE_CHANGE + MANDATORY_PER_SESSION))
+    missing = [t for t in all_mandatory if t not in run]
+
+    print("\n─── SESSION END AUDIT ───────────────────────────────")
+    print(f"Teams run: {run or 'none'}")
+
+    if missing:
+        print(f"\n⚠️  MISSED THIS SESSION:")
+        for t in missing:
+            print(f"   ❌ {t}")
+        print("\nStore context before ending:")
+        print("  python3 execution/memory_manager.py store --content '...' --type decision --project <project>")
+        print("  python3 execution/cross_agent_context.py broadcast --agent claude --message '...' --project <project>")
+    else:
+        print("✅ All mandatory steps completed this session.")
+    print("─────────────────────────────────────────────────────\n")
+
+    # Always run memory proof
+    os.system("python3 execution/memory_usage_proof.py --check --since 120 2>/dev/null | tail -3")
+
+
+def reset():
+    state = {"teams_run": [], "session_start": datetime.now().isoformat(), "commits_since_review": 0}
+    save_state(state)
+    print("✅ Session checklist reset.")
+
+
+if __name__ == "__main__":
+    cmd = sys.argv[1] if len(sys.argv) > 1 else "check-commit"
+    if cmd == "check-commit":
+        check_commit()
+    elif cmd == "mark" and len(sys.argv) > 2:
+        mark(sys.argv[2])
+    elif cmd == "session-end":
+        session_end()
+    elif cmd == "reset":
+        reset()
+    else:
+        print(f"Unknown command: {cmd}")
+        sys.exit(1)