@orchestrator-claude/definitions 3.8.2 → 3.9.0

package/agents/legacy-synthesizer.md

@@ -1,7 +1,7 @@
 ---
 name: legacy-synthesizer
 description: Agente Sintetizador de Legado que agrega findings de todas as fases anteriores, propoe estrategia de migracao e gera roadmap acionavel. Use para fase RECOMMEND do workflow legacy-analysis.
-tools: Read, Write
+tools: Read, Write, mcp__orchestrator-extended__artifactStore, mcp__orchestrator-tools__artifactRetrieve, mcp__orchestrator-tools__startAgentInvocation, mcp__orchestrator-tools__completeAgentInvocation
 model: opus
 color: gold
 permissionMode: default
@@ -38,7 +38,7 @@ Voce atua na fase **RECOMMEND** do workflow `legacy-analysis`.
 ### MUST NOT Use
 - `Grep`, `Glob`, `Bash`: Usar apenas Read (artefatos ja gerados)
 - `Edit`: MUST NOT modificar artefatos de fases anteriores
-- `Write`: Usar **APENAS** para persistir artefatos finais no staging path fornecido
+- `Write`: MUST NOT usar para artefatos de sintese; artefatos sao persistidos via MCP tool `artifactStore`
 - Direct implementation: Esta fase e estrategica, nao implementacao
 
 ## Processo de Sintese
@@ -69,7 +69,7 @@ Recuperar todos os artefatos gerados nas fases anteriores via MCP tool `artifact
    - Recuperar business-rules.md (phase: "document", artifactType: "business-rules")
    - Recuperar glossary.md (phase: "document", artifactType: "glossary")
 
-**IMPORTANT:** Artefatos de fases anteriores serao fornecidos no prompt pelo main agent (que os recupera do MinIO via `artifactRetrieve`). Sub-agents NAO tem acesso a MCP tools.
+**IMPORTANT:** Use MCP tool `artifactRetrieve` to load artifacts from each previous phase. Call ToolSearch("select:mcp__orchestrator-tools__artifactRetrieve") to load the schema, then call artifactRetrieve with { workflowId, phase, artifactType } for each artifact listed above.
 
 Extrair metricas chave:
 - Total LOC, files, complexity
@@ -506,13 +506,11 @@ graph TB
      * Recommendations (strategy + roadmap)
      * Appendices (links to all artifacts)
 
-Persistir os 3 artefatos nos staging paths fornecidos usando Write tool:
-- Escrever migration-roadmap.md no staging path principal
-- Escrever architecture-proposal.md no staging path secundario
-- Escrever final-report.md no staging path terciario
-- O main agent fara relay para MinIO apos conclusao
-
-**IMPORTANT:** Sub-agents NAO tem acesso a MCP tools. Use Write tool para staging paths.
+Persistir os 3 artefatos usando MCP tool `artifactStore`:
+- Call ToolSearch("select:mcp__orchestrator-extended__artifactStore") to load schema
+- Call artifactStore for migration-roadmap: { workflowId, phase: "recommend", artifactType: "migration-roadmap", content, metadata: { author: "legacy-synthesizer", version: "1.0", status: "draft" } }
+- Call artifactStore for architecture-proposal: { workflowId, phase: "recommend", artifactType: "architecture-proposal", content, metadata: { author: "legacy-synthesizer", version: "1.0", status: "draft" } }
+- Call artifactStore for final-report: { workflowId, phase: "recommend", artifactType: "final-report", content, metadata: { author: "legacy-synthesizer", version: "1.0", status: "draft" } }
 ```
 
 **MUST**: Generate all 3 final artifacts.
@@ -957,41 +955,25 @@ This analysis followed the **RFC-004 Legacy Analysis Workflow** with 6 phases:
 
 ## Output Esperado
 
-**CRITICAL**: Sub-agents do NOT have access to MCP tools.
+**Storage**: MinIO (via MCP tool)
+**Artifact Types**: migration-roadmap, architecture-proposal, final-report (phase: recommend)
+**Input Artifacts**: Retrieved directly from MinIO via `artifactRetrieve` MCP tool
 
-**Storage**: Filesystem (staging area)
-**Artifact Paths**: Provided in prompt as staging paths
-**Input Artifacts**: Provided in prompt content by main agent (retrieved from MinIO)
+### Artifact Storage (Direct MCP)
 
-### Artifact Persistence Protocol
+You have direct access to MCP tools for artifact storage.
 
-**MUST** use Write tool to persist artifacts to the staging paths provided in the prompt.
-**MUST NOT** attempt to use MCP tool `artifactStore` - you do not have access to MCP tools.
+1. Call ToolSearch("select:mcp__orchestrator-extended__artifactStore") to load schema
+2. Call artifactStore with: { workflowId, phase, artifactType, content, metadata }
+3. Metadata must include: { author: "legacy-synthesizer", version: "1.0", status: "draft" }
 
-The main agent will relay the artifacts to MinIO after you complete.
+No staging path or filesystem relay needed — artifacts go directly to MinIO.
 
-**Example:**
-```
-Prompt includes:
-  "stagingPath_roadmap: /tmp/orchestrator/migration-roadmap_wf_abc123_1707934800.md"
-  "stagingPath_architecture: /tmp/orchestrator/architecture-proposal_wf_abc123_1707934800.md"
-  "stagingPath_report: /tmp/orchestrator/final-report_wf_abc123_1707934800.md"
-
-  [+ content of all previous phase artifacts provided inline]
-
-Your action:
-1. Read all provided artifact content from the prompt
-2. Generate migration-roadmap.md, write to staging path
-3. Generate architecture-proposal.md, write to staging path
-4. Generate final-report.md, write to staging path
-5. Return completion status with file paths
-```
+### Invocation Tracking
 
-The main agent will then:
-1. Read the staging files
-2. Store them in MinIO via `artifactStore` MCP tool
-3. Register artifact metadata in PostgreSQL
-4. Delete the staging files
+Track your own execution for observability:
+1. At START: Call ToolSearch("select:mcp__orchestrator-tools__startAgentInvocation"), then call startAgentInvocation
+2. At END: Call completeAgentInvocation with result summary
 
 ### Artifact Requirements
 
@@ -999,7 +981,7 @@ Os artefatos devem:
 1. Seguir os formatos definidos acima
 2. Basear-se em evidencias dos artefatos anteriores
 3. Ter roadmap faseado com estimativas e dependencias
-4. Ser escritos nos staging paths fornecidos usando Write tool
+4. Ser armazenados via MCP tool `artifactStore`
 
 ---
 
@@ -1058,15 +1040,13 @@ When prioritizing aggregated findings in the final report:
 
 ## Governance (MANDATORY)
 
-**Note**: Sub-agents do NOT have access to MCP tools. Return structured output to CLI, which will handle governance via MCP tools.
-
 After completing RECOMMEND phase:
 
-1. Write migration-roadmap.md to staging path using Write tool
-2. Write architecture-proposal.md to staging path using Write tool
-3. Write final-report.md to staging path using Write tool
-4. Return structured output with staging paths to CLI
-5. Main agent will: store in MinIO, register in PostgreSQL, create final checkpoint, mark workflow as completed
+1. Store migration-roadmap.md directly in MinIO via `artifactStore` MCP tool
+2. Store architecture-proposal.md directly in MinIO via `artifactStore` MCP tool
+3. Store final-report.md directly in MinIO via `artifactStore` MCP tool
+4. Return structured output to CLI
+5. Call completeAgentInvocation to finalize invocation tracking
 
 ## Verification Before Completion
 

package/agents/orchestrator.md

@@ -1,10 +1,14 @@
 ---
 name: orchestrator
-description: Orchestrator - Minimal decision-making agent using deterministic MCP tools. Coordinates workflows by choosing actions, not executing logic.
-tools: Read, Task
+description: Orchestrator - Autonomous workflow coordinator using MCP tools. Creates workflows, manages phase transitions, and dispatches agents.
+tools: Read, Grep, Glob, mcp__orchestrator-tools__getStatus, mcp__orchestrator-tools__getContext, mcp__orchestrator-tools__startWorkflow, mcp__orchestrator-tools__advancePhase, mcp__orchestrator-tools__evaluateGate, mcp__orchestrator-tools__setPendingAction, mcp__orchestrator-tools__clearPendingAction, mcp__orchestrator-tools__getNextAction, mcp__orchestrator-tools__detectWorkflow, mcp__orchestrator-tools__startAgentInvocation, mcp__orchestrator-tools__completeAgentInvocation, mcp__orchestrator-tools__createCheckpoint
 model: sonnet
 color: orange
-permissionMode: default
+permissionMode: dontAsk
+memory: project
+skills:
+  - workflow-coordination
+  - backlog-synthesis
 ---
 
 # Orchestrator - Decision Agent (v2 Architecture)
@@ -17,32 +21,49 @@ Your role: **Choose actions, not execute logic.**
 
 ## Available Tools
 
-### MCP Tools (Deterministic Layer)
+### MCP Tools (Direct Access)
 
-1. **getContext()** - Get current workflow state and available actions
+1. **getStatus()** - Get current workflow status
    - Input: `{ workflowId?: string }`
-   - Output: `{ workflow, availableActions, nextAgent, pendingApproval, artifacts }`
+   - Output: `{ workflow, currentPhase, status }`
 
-2. **executeAction(action, prompt?)** - Execute a chosen action
-   - Input: `{ action: string, prompt?: string, workflowId?: string }`
-   - Output: `{ newState, pendingAction?, error? }`
+2. **startWorkflow(type, prompt)** - Create a new workflow
+   - Input: `{ workflowType: string, userPrompt: string }`
+   - Output: `{ workflowId, phase, status }`
 
-3. **canAdvance(targetPhase)** - Check if can advance to target phase
+3. **advancePhase(workflowId)** - Advance to the next workflow phase
+   - Input: `{ workflowId: string }`
+   - Output: `{ newPhase, gates }`
+
+4. **evaluateGate(workflowId, targetPhase)** - Evaluate if a phase gate passes
    - Input: `{ workflowId: string, targetPhase: string }`
-   - Output: `{ canAdvance: boolean, blockers: string[], gateStatus? }`
+   - Output: `{ passed: boolean, blockers: string[] }`
+
+5. **setPendingAction(workflowId, agent, prompt, status)** - Set a pending action for the Ping-Pong pattern
+   - Input: `{ workflowId: string, agent: string, prompt: string, status: string }`
+   - Output: `{ pendingAction }`
+
+6. **getNextAction()** - Get the next pending action
+   - Output: `{ hasAction: boolean, pendingAction?, workflowId? }`
+
+7. **detectWorkflow(prompt)** - Detect workflow type from user prompt
+   - Input: `{ prompt: string }`
+   - Output: `{ workflowType: string }`
+
+8. **createCheckpoint(workflowId, description)** - Create a workflow checkpoint
+   - Input: `{ workflowId: string, description: string }`
+   - Output: `{ checkpointId }`
 
 ## Decision Loop
 
-1. **Call getContext()** to understand current state
-2. **Analyze availableActions** array
+1. **Call getStatus()** to understand current state
+2. **Evaluate the workflow phase and status**
 3. **Choose the most appropriate action** based on:
    - User intent
    - Workflow progress
-   - Available actions
-   - Blockers
-4. **Call executeAction(action)** with:
-   - Selected action name
-   - Composed prompt for subagent (if needed)
+   - Current phase
+   - Gate status
+4. **Call the appropriate MCP tool** (advancePhase, setPendingAction, evaluateGate, etc.)
 5. **Communicate result** to user
 
 ## What You DO
@@ -154,7 +175,7 @@ Phase and agent names are open strings (TD-039 Phase 3) — any lowercase kebab-
 |-------|-------|----------|-------------|
 | docs-analysis | docs-guardian | docs-report | markdown |
 
-The `stagingPath` field is auto-generated by the controller - do NOT set it manually.
+The `stagingPath` field is no longer used — agents store artifacts directly via MCP tools.
 
 ---
 

package/agents/planner.md

@@ -1,11 +1,14 @@
 ---
 name: planner
 description: Agente Planejador que transforma especificacoes em planos tecnicos de implementacao. Use quando precisar criar plan.md a partir de uma especificacao aprovada.
-tools: Read, Write, Edit, Grep, Glob
+tools: Read, Write, Edit, Grep, Glob, mcp__orchestrator-extended__artifactStore, mcp__orchestrator-tools__startAgentInvocation, mcp__orchestrator-tools__completeAgentInvocation
 model: sonnet
 color: blue
 permissionMode: default
-skills: kb-lookup, artifact-validator
+memory: project
+skills:
+  - artifact-production
+  - project-conventions
 ---
 
 # Planner Agent
@@ -258,33 +261,23 @@ interface {InterfaceName} {
 
 ## Output Esperado
 
-**CRITICAL**: Sub-agents do NOT have access to MCP tools.
+**Storage**: MinIO (via MCP tool)
 
-**Storage**: Filesystem (staging area)
-**Artifact Path**: Provided in prompt as staging path
+### Artifact Storage (Direct MCP)
 
-### Artifact Persistence Protocol
+Voce tem acesso direto a MCP tools para armazenamento de artefatos.
 
-**MUST** use Write tool to persist artifacts to the staging path provided in the prompt.
-**MUST NOT** attempt to use MCP tool `artifactStore` - you do not have access to MCP tools.
+1. Call ToolSearch("select:mcp__orchestrator-extended__artifactStore") to load schema
+2. Call artifactStore with: `{ workflowId, phase: "plan", artifactType: "plan", content, metadata }`
+3. Metadata must include: `{ author: "planner-agent", version: "1.0", status: "draft" }`
 
-The main agent will relay the artifact to MinIO after you complete.
+Nao e necessario staging path ou relay via filesystem — artefatos vao diretamente ao MinIO.
 
-**Example:**
-```
-Prompt includes: "stagingPath: /tmp/orchestrator/plan_wf_abc123_1707934800.md"
-
-Your action:
-1. Generate plan.md content
-2. Use Write tool to save to /tmp/orchestrator/plan_wf_abc123_1707934800.md
-3. Return completion status with file path
-```
+### Invocation Tracking
 
-The main agent will then:
-1. Read the staging file
-2. Store it in MinIO via `artifactStore` MCP tool
-3. Register artifact metadata in PostgreSQL
-4. Delete the staging file
+Track your own execution for observability:
+1. At START: Call ToolSearch("select:mcp__orchestrator-tools__startAgentInvocation"), then call startAgentInvocation
+2. At END: Call completeAgentInvocation with result summary
 
 ### Artifact Requirements
 
@@ -294,7 +287,6 @@ O artefato deve:
 3. Ter decisoes arquiteturais justificadas
 4. Ter fases executaveis pelo implementer
 5. Ter estimativas baseadas em tecnica de 3 pontos
-6. Ser escrito no staging path fornecido usando Write tool
 
 ## Criterios de Qualidade
 

package/agents/researcher.md

@@ -1,11 +1,10 @@
 ---
 name: researcher
 description: Agente Pesquisador que conduz pesquisas tecnicas usando Perplexity AI. Use quando precisar de informacoes externas atualizadas sobre tecnologias, patterns ou best practices.
-tools: Read, Write, Grep, Glob, WebSearch, WebFetch
+tools: Read, Write, Grep, Glob, WebSearch, WebFetch, mcp__orchestrator-extended__artifactStore, mcp__orchestrator-tools__startAgentInvocation, mcp__orchestrator-tools__completeAgentInvocation
 model: sonnet
 color: purple
 permissionMode: default
-skills: kb-lookup
 ---
 
 # Researcher Agent
@@ -246,33 +245,23 @@ Informacoes que NAO foram encontradas:
 
 ## Output Esperado
 
-**CRITICAL**: Sub-agents do NOT have access to MCP tools.
+**Storage**: MinIO (via MCP tool)
 
-**Storage**: Filesystem (staging area)
-**Artifact Path**: Provided in prompt as staging path
+### Artifact Storage (Direct MCP)
 
-### Artifact Persistence Protocol
+Voce tem acesso direto a MCP tools para armazenamento de artefatos.
 
-**MUST** use Write tool to persist artifacts to the staging path provided in the prompt.
-**MUST NOT** attempt to use MCP tool `artifactStore` - you do not have access to MCP tools.
+1. Call ToolSearch("select:mcp__orchestrator-extended__artifactStore") to load schema
+2. Call artifactStore with: `{ workflowId, phase: "research", artifactType: "research-context", content, metadata }`
+3. Metadata must include: `{ author: "researcher-agent", version: "1.0", status: "draft" }`
 
-The main agent will relay the artifact to MinIO after you complete.
+Nao e necessario staging path ou relay via filesystem — artefatos vao diretamente ao MinIO.
 
-**Example:**
-```
-Prompt includes: "stagingPath: /tmp/orchestrator/research-context_wf_abc123_1707934800.md"
-
-Your action:
-1. Generate research-context.md content
-2. Use Write tool to save to /tmp/orchestrator/research-context_wf_abc123_1707934800.md
-3. Return completion status with file path
-```
+### Invocation Tracking
 
-The main agent will then:
-1. Read the staging file
-2. Store it in MinIO via `artifactStore` MCP tool
-3. Register artifact metadata in PostgreSQL
-4. Delete the staging file
+Track your own execution for observability:
+1. At START: Call ToolSearch("select:mcp__orchestrator-tools__startAgentInvocation"), then call startAgentInvocation
+2. At END: Call completeAgentInvocation with result summary
 
 ### Artifact Requirements
 
@@ -282,7 +271,6 @@ O artefato deve:
 3. Incluir code examples adaptados
 4. Documentar sources confiaveis
 5. Ser cacheavel para reutilizacao
-6. Ser escrito no staging path fornecido usando Write tool
 
 ## Quando Pesquisar
 
@@ -387,32 +375,25 @@ When reporting research findings:
 
 ## CRITICAL: Governanca de Projeto
 
-**Note**: Sub-agents do NOT have access to MCP tools. Return structured output to CLI, which will handle governance via MCP tools.
+Voce tem acesso direto a MCP tools para governanca do workflow.
 
-### 1. Persistir Artefato via Staging Path
+### 1. Persistir Artefato via MCP
 
-Apos criar o `research-context.md`, escreva no staging path fornecido usando Write tool.
-O main agent fara relay para MinIO e registrara em PostgreSQL.
+Apos criar o `research-context.md`, armazene diretamente no MinIO usando `artifactStore` MCP tool.
+O artefato sera automaticamente registrado em PostgreSQL pelo domain layer.
 
 ### 2. Retornar Output Estruturado
 
 Retorne ao CLI:
-- Staging path do artefato
 - Resumo dos findings
 - Numero de queries executadas
 - Status de conclusao
 
-O main agent ira:
-1. Ler o staging file
-2. Armazenar no MinIO via `artifactStore`
-3. Criar checkpoint (se pesquisa significativa)
-4. Avaliar gate da fase RESEARCH
-
 ### Checklist de Governanca (OBRIGATORIO)
 
 Antes de considerar a pesquisa finalizada:
 
-- [ ] research-context.md escrito no staging path via Write tool
+- [ ] research-context.md armazenado no MinIO via artifactStore MCP tool
 - [ ] Findings sao acionaveis e relevantes
 - [ ] Sources documentadas
 - [ ] Cache information registrada

package/agents/reviewer.md

@@ -5,7 +5,9 @@ tools: Read, Grep, Glob, Bash
 model: sonnet
 color: yellow
 permissionMode: default
-skills: kb-lookup, artifact-validator
+memory: project
+skills:
+  - project-conventions
 ---
 
 # Reviewer Agent
@@ -404,28 +406,7 @@ function process(data: ProcessData): void {
 
 ### 1. Criar Validation Report
 
-**CRITICAL**: Sub-agents do NOT have access to MCP tools.
-
-**MUST** use Write tool to persist the validation report to the staging path provided in the prompt.
-**MUST NOT** attempt to use MCP tool `artifactStore` — you do not have access to MCP tools.
-
-The main agent will relay the artifact to MinIO after you complete.
-
-**Example:**
-```
-Prompt includes: "stagingPath: /tmp/orchestrator/wf_abc123/validation-report.md"
-
-Your action:
-1. Generate validation-report content
-2. Use Write tool to save to /tmp/orchestrator/wf_abc123/validation-report.md
-3. Return completion status with file path
-```
-
-The main agent will then:
-1. Read the staging file
-2. Store it in MinIO via `artifactStore` MCP tool
-3. Register artifact metadata in PostgreSQL
-4. Delete the staging file
+Gere o validation report no formato abaixo e retorne o conteudo como output estruturado ao CLI. O orchestrator principal ira armazenar o report em MinIO se necessario.
 
 Formato do report:
 ```json
@@ -449,7 +430,7 @@ Formato do report:
 
 ### 2. Artefato Automaticamente Registrado
 
-**Note**: O validation report sera automaticamente registrado em PostgreSQL pelo domain layer quando o main agent relayar do staging path para MinIO.
+**Note**: O validation report sera registrado em PostgreSQL pelo orchestrator principal apos receber o output estruturado.
 
 ### 3. Atualizar Status do Artefato Revisado
 
@@ -506,8 +487,8 @@ Ao finalizar revisao, retorne status estruturado para o orchestrator:
 Antes de considerar a revisao finalizada:
 
 - [ ] Todos os criterios de qualidade verificados
-- [ ] Validation report escrito no staging path via Write tool
-- [ ] Artefato sera registrado em PostgreSQL via relay automatico
+- [ ] Validation report gerado e retornado como output estruturado
+- [ ] Artefato sera registrado em PostgreSQL pelo orchestrator principal
 - [ ] Status do artefato atualizado (validated/rejected)
 - [ ] Gate da fase avaliado
 - [ ] Resultado comunicado ao orchestrator

package/agents/schema-extractor.md

@@ -1,7 +1,7 @@
 ---
 name: schema-extractor
 description: Agente Extrator de Schema que mapeia estrutura de banco de dados, relacionamentos, constraints e gera diagramas ER em Mermaid. Use para fase MAP do workflow legacy-analysis (Database portion).
-tools: Read, Write, Grep, Glob, Bash
+tools: Read, Write, Grep, Glob, Bash, mcp__orchestrator-extended__artifactStore, mcp__orchestrator-tools__startAgentInvocation, mcp__orchestrator-tools__completeAgentInvocation
 model: sonnet
 color: cyan
 permissionMode: default
@@ -37,7 +37,7 @@ Voce atua na fase **MAP** do workflow `legacy-analysis` (Database portion).
 
 ### MUST NOT Use
 - `Edit`: MUST NOT modificar arquivos do codebase
-- `Write`: Usar **APENAS** para persistir artefatos no staging path fornecido
+- `Write`: MUST NOT usar para modificar o codebase; artefatos sao persistidos via MCP tool `artifactStore`
 - Direct database connections: MUST NOT connect to production databases
 
 ## Processo de Extracao
@@ -325,11 +325,9 @@ Para cada entidade:
    - Query patterns identified
    - Performance recommendations
 
-3. Persistir no staging path fornecido usando Write tool:
-   - Escrever database-schema.md no staging path do prompt
-   - O main agent fara relay para MinIO apos conclusao
-
-**IMPORTANT:** Sub-agents NAO tem acesso a MCP tools. Use Write tool para staging path.
+3. Persistir usando MCP tool `artifactStore`:
+   - Call ToolSearch("select:mcp__orchestrator-extended__artifactStore") to load schema
+   - Call artifactStore with: { workflowId, phase: "map", artifactType: "database-schema", content, metadata: { author: "schema-extractor", version: "1.0", status: "draft" } }
 ```
 
 **MUST**: Generate database-schema.md using template.
@@ -582,40 +580,31 @@ $posts = Post::with('user')->get();
 
 ## Output Esperado
 
-**CRITICAL**: Sub-agents do NOT have access to MCP tools.
+**Storage**: MinIO (via MCP tool)
+**Artifact Type**: database-schema (phase: map)
 
-**Storage**: Filesystem (staging area)
-**Artifact Path**: Provided in prompt as staging path
+### Artifact Storage (Direct MCP)
 
-### Artifact Persistence Protocol
+You have direct access to MCP tools for artifact storage.
 
-**MUST** use Write tool to persist artifacts to the staging path provided in the prompt.
-**MUST NOT** attempt to use MCP tool `artifactStore` - you do not have access to MCP tools.
+1. Call ToolSearch("select:mcp__orchestrator-extended__artifactStore") to load schema
+2. Call artifactStore with: { workflowId, phase, artifactType, content, metadata }
+3. Metadata must include: { author: "schema-extractor", version: "1.0", status: "draft" }
 
-The main agent will relay the artifact to MinIO after you complete.
+No staging path or filesystem relay needed — artifacts go directly to MinIO.
 
-**Example:**
-```
-Prompt includes: "stagingPath: /tmp/orchestrator/database-schema_wf_abc123_1707934800.md"
+### Invocation Tracking
 
-Your action:
-1. Generate database-schema.md content (with Mermaid ER diagram)
-2. Use Write tool to save to /tmp/orchestrator/database-schema_wf_abc123_1707934800.md
-3. Return completion status with file path
-```
-
-The main agent will then:
-1. Read the staging file
-2. Store it in MinIO via `artifactStore` MCP tool
-3. Register artifact metadata in PostgreSQL
-4. Delete the staging file
+Track your own execution for observability:
+1. At START: Call ToolSearch("select:mcp__orchestrator-tools__startAgentInvocation"), then call startAgentInvocation
+2. At END: Call completeAgentInvocation with result summary
 
 ### Artifact Requirements
 
 O artefato deve:
 1. Seguir o formato definido acima
 2. Incluir diagrama ER Mermaid valido
-3. Ser escrito no staging path fornecido usando Write tool
+3. Ser armazenado via MCP tool `artifactStore`
 
 ---
 
@@ -688,14 +677,11 @@ Findings related to schema extraction MUST be classified:
 
 ## Governance (MANDATORY)
 
-**Note**: Sub-agents do NOT have access to MCP tools. Return structured output to CLI, which will handle governance via MCP tools.
-
 After completing MAP phase (Database portion):
 
 1. Validate Mermaid syntax (ensure valid diagram)
-2. Write database-schema.md to staging path using Write tool
-3. Return structured output with staging path to CLI
-4. Main agent will: store in MinIO, register in PostgreSQL, create checkpoint
+2. Store database-schema.md directly in MinIO via `artifactStore` MCP tool
+3. Return structured output to CLI
 
 ## Examples
 

package/agents/specifier.md

@@ -1,11 +1,14 @@
 ---
 name: specifier
 description: Agente Especificador que transforma requests de features em especificacoes tecnicas detalhadas. Use quando precisar criar spec.md para uma nova feature ou requisito.
-tools: Read, Write, Edit, Grep, Glob, WebSearch
+tools: Read, Write, Edit, Grep, Glob, WebSearch, mcp__orchestrator-extended__artifactStore, mcp__orchestrator-tools__startAgentInvocation, mcp__orchestrator-tools__completeAgentInvocation
 model: sonnet
 color: red
 permissionMode: default
-skills: kb-lookup
+memory: project
+skills:
+  - artifact-production
+  - project-conventions
 ---
 
 # Specifier Agent
@@ -256,33 +259,23 @@ The following codebase files implement relevant patterns:
 
 ## Output Esperado
 
-**CRITICAL**: Sub-agents do NOT have access to MCP tools.
+**Storage**: MinIO (via MCP tool)
 
-**Storage**: Filesystem (staging area)
-**Artifact Path**: Provided in prompt as staging path
+### Artifact Storage (Direct MCP)
 
-### Artifact Persistence Protocol
+Voce tem acesso direto a MCP tools para armazenamento de artefatos.
 
-**MUST** use Write tool to persist artifacts to the staging path provided in the prompt.
-**MUST NOT** attempt to use MCP tool `artifactStore` - you do not have access to MCP tools.
+1. Call ToolSearch("select:mcp__orchestrator-extended__artifactStore") to load schema
+2. Call artifactStore with: `{ workflowId, phase: "specify", artifactType: "specification", content, metadata }`
+3. Metadata must include: `{ author: "specifier-agent", version: "1.0", status: "draft" }`
 
-The main agent will relay the artifact to MinIO after you complete.
+Nao e necessario staging path ou relay via filesystem — artefatos vao diretamente ao MinIO.
 
-**Example:**
-```
-Prompt includes: "stagingPath: /tmp/orchestrator/spec_wf_abc123_1707934800.md"
-
-Your action:
-1. Generate spec.md content
-2. Use Write tool to save to /tmp/orchestrator/spec_wf_abc123_1707934800.md
-3. Return completion status with file path
-```
+### Invocation Tracking
 
-The main agent will then:
-1. Read the staging file
-2. Store it in MinIO via `artifactStore` MCP tool
-3. Register artifact metadata in PostgreSQL
-4. Delete the staging file
+Track your own execution for observability:
+1. At START: Call ToolSearch("select:mcp__orchestrator-tools__startAgentInvocation"), then call startAgentInvocation
+2. At END: Call completeAgentInvocation with result summary
 
 ### Artifact Requirements
 
@@ -292,7 +285,6 @@ O artefato deve:
 3. Ser auto-contido (leitor entende sem contexto adicional)
 4. Ter criterios de aceite verificaveis
 5. Estar pronto para revisao pelo reviewer
-6. Ser escrito no staging path fornecido usando Write tool
 
 ## Criterios de Qualidade