aiox-core 5.0.3 → 5.0.4

package/pro/squads/squad-creator-pro/tasks/qualify-task.md

@@ -1,265 +1,265 @@
-# Task: qualify-task
-
-> **Model Tier Qualification** | Testar empiricamente se task pode rodar em Haiku
-
-**Task ID:** qualify-task
-**Version:** 1.0.0
-**Command:** `*qualify-task <task-name> [--batch]`
-**Execution Type:** Workflow Orchestrator
-**Model:** `Haiku` (orchestration only - workflow handles model selection)
-**Haiku Eligible:** YES — orchestration is deterministic
-**Orchestrator:** @pedro-valerio
-
-## Purpose
-
-Invocar `wf-model-tier-qualification.yaml` para testar empiricamente se uma task pode rodar em modelo mais barato (Haiku) mantendo qualidade.
-
-**Diferença do `*optimize`:**
-- `*optimize --hybrid` → Pipeline completo (análise + conversão + qualificação + GAP ZERO)
-- `*qualify-task` → APENAS qualificação empírica (Opus vs Haiku)
-
-## Usage
-
-```bash
-# Qualificar uma task específica
-*qualify-task pv-axioma-assessment
-
-# Qualificar com threshold customizado (default: 90%)
-*qualify-task validate-squad --threshold 0.85
-
-# Qualificar batch (todas tasks não-qualificadas)
-*qualify-task --batch
-
-# Qualificar batch de uma wave específica
-*qualify-task --batch --wave 1
-```
-
-## Veto Conditions
-
-- [ ] VETO se task não existe em `squads/squad-creator-pro/tasks/`
-- [ ] VETO se task não tem `test_input` no registry do workflow
-- [ ] VETO se já está qualificada em `model-routing.yaml` (use `--force` para re-testar)
-
----
-
-## Workflow
-
-### Step 1: Pre-flight Validation
-
-```yaml
-preflight:
-  - check: "Task file exists?"
-    path: "squads/squad-creator-pro/tasks/{task_name}.md"
-    veto_if: "not exists"
-
-  - check: "Task in test_input_registry?"
-    lookup: "wf-model-tier-qualification.yaml → test_input_registry.{task_name}"
-    veto_if: "not found"
-    message: "Task needs test_input defined first. Add to workflow registry."
-
-  - check: "Already qualified?"
-    lookup: "config/model-routing.yaml → tasks.{task_name}.validated"
-    action_if_true: "Skip unless --force flag"
-```
-
-### Step 2: Load Workflow
-
-```yaml
-load_workflow:
-  file: "squads/squad-creator-pro/workflows/wf-model-tier-qualification.yaml"
-  action: "Read COMPLETE workflow definition"
-
-  extract:
-    - phases (4 phases + auto-improve loop)
-    - test_input_registry
-    - veto_conditions
-    - compensations
-```
-
-### Step 3: Execute Qualification
-
-```yaml
-execute:
-  # Invoke workflow with inputs
-  inputs:
-    task_name: "{from command}"
-    target_tier: "haiku"  # default
-    threshold: 0.90       # default, can override with --threshold
-
-  phases:
-    - phase_0: "PRE-FLIGHT"
-    - phase_1: "PARALLEL EXECUTION (Opus + Haiku)"
-    - phase_2: "COMPARE & AUTO-DECIDE"
-    - phase_3: "AUTO-IMPROVE LOOP (if needed)"
-    - phase_4: "GENERATE REPORT"
-
-  # Follow workflow phases exactly as defined
-  action: |
-    Execute wf-model-tier-qualification.yaml with the provided inputs.
-    The workflow handles:
-    - Running Opus baseline
-    - Running Haiku candidate
-    - Comparing results
-    - Applying compensations if needed
-    - Generating qualification report
-```
-
-### Step 4: Update Config
-
-```yaml
-update_config:
-  condition: "decision is QUALIFIED or QUALIFIED_WITH_COMPENSATION"
-
-  action: |
-    Update config/model-routing.yaml:
-
-    tasks:
-      {task_name}.md:
-        tier: haiku  # or sonnet if escalated
-        confidence: high
-        validated: true
-        test_date: "{today}"
-        test_result: "{summary from qualification report}"
-
-  script: |
-    node squads/squad-creator-pro/scripts/model-tier-validator.cjs \
-      update-routing {task_name} {recommended_tier} "{reason}"
-```
-
-### Step 5: Summary
-
-```yaml
-summary:
-  display: |
-    ╔══════════════════════════════════════════════════════════════════╗
-    ║  QUALIFICATION RESULT: {task_name}                               ║
-    ╠══════════════════════════════════════════════════════════════════╣
-    ║                                                                  ║
-    ║  Decision: {QUALIFIED | NEEDS_COMPENSATION | ESCALATE | OPUS}    ║
-    ║  Tier: {haiku | sonnet | opus}                                   ║
-    ║  Quality: {pct}% of Opus baseline                                ║
-    ║  Cost Savings: {savings_pct}%                                    ║
-    ║                                                                  ║
-    ║  Report: test-cases/{task_name}/qualification-report.yaml        ║
-    ╚══════════════════════════════════════════════════════════════════╝
-```
-
----
-
-## Batch Mode
-
-```yaml
-batch_mode:
-  trigger: "--batch flag"
-
-  action: |
-    1. Read wf-model-tier-qualification.yaml → batch_mode.execution_order
-    2. Filter tasks not yet validated in model-routing.yaml
-    3. Execute qualification for each task sequentially
-    4. Generate consolidated QUALIFICATION-DASHBOARD.yaml
-
-  waves:
-    wave_1: "High-confidence Haiku candidates"
-    wave_2: "Admin tasks (script-based)"
-    wave_3: "Sonnet candidates tested with Haiku"
-
-  output:
-    - "test-cases/{task}/qualification-report.yaml" (per task)
-    - "test-cases/QUALIFICATION-DASHBOARD.yaml" (consolidated)
-```
-
----
-
-## Examples
-
-### Example 1: Qualify Single Task
-
-```
-User: *qualify-task an-assess-sources
-
-Agent:
-## Step 1: Pre-flight ✅
-- Task file: squads/squad-creator-pro/tasks/an-assess-sources.md ✅
-- Test input: outputs/minds/pedro_valerio/sources/ ✅
-- Already qualified: NO
-
-## Step 2: Loading workflow...
-Read: wf-model-tier-qualification.yaml
-
-## Step 3: Executing qualification...
-[Phase 0: PRE-FLIGHT] ✅
-[Phase 1: PARALLEL EXECUTION]
-  - Opus baseline running...
-  - Haiku candidate running...
-[Phase 2: COMPARE]
-  - Opus: 8.5/10
-  - Haiku: 8.2/10 (96.5% quality)
-  - Decision: QUALIFIED ✅
-
-## Step 4: Updating config...
-Updated model-routing.yaml: an-assess-sources.md → haiku
-
-## Result
-╔══════════════════════════════════════════════════════════════════╗
-║  QUALIFICATION RESULT: an-assess-sources                         ║
-╠══════════════════════════════════════════════════════════════════╣
-║  Decision: QUALIFIED                                             ║
-║  Tier: haiku                                                     ║
-║  Quality: 96.5%                                                  ║
-║  Savings: 97%                                                    ║
-╚══════════════════════════════════════════════════════════════════╝
-```
-
-### Example 2: Batch Mode
-
-```
-User: *qualify-task --batch --wave 1
-
-Agent:
-## Batch Qualification: Wave 1 (10 tasks)
-
-| # | Task | Status | Tier | Quality | Savings |
-|---|------|--------|------|---------|---------|
-| 1 | pv-axioma-assessment | ✅ QUALIFIED | haiku | 95.5% | 97% |
-| 2 | validate-squad | ✅ QUALIFIED | haiku | 91% | 98% |
-| 3 | an-fidelity-score | ✅ SCRIPT | script | 100% | 100% |
-| 4 | an-diagnose-clone | ❌ OPUS REQUIRED | opus | 30% | 0% |
-...
-
-## Summary
-- Qualified: 8/10 (80%)
-- Script-only: 2/10 (20%)
-- Opus required: 1/10 (10%)
-- Total savings: ~85% token cost reduction
-
-Dashboard: test-cases/QUALIFICATION-DASHBOARD.yaml
-```
-
----
-
-## Completion Criteria
-
-| Critério | Status |
-|----------|--------|
-| Pre-flight validation passed | [ ] |
-| Workflow executed completely | [ ] |
-| Qualification report generated | [ ] |
-| Config updated (if qualified) | [ ] |
-| Summary displayed | [ ] |
-
----
-
-## Related
-
-- `wf-model-tier-qualification.yaml` - The actual workflow being invoked
-- `config/model-routing.yaml` - Where results are stored
-- `*optimize --hybrid` - Full pipeline (includes qualification)
-- `*lookup-model` - Read-only lookup (no qualification)
-- `*smoke-test-model-routing` - Test the routing system itself
-
----
-
-_Task Version: 1.0.0_
-_Last Updated: 2026-02-12_
-
-*"Não assumir. Provar com dados."*
+# Task: qualify-task
+
+> **Model Tier Qualification** | Testar empiricamente se task pode rodar em Haiku
+
+**Task ID:** qualify-task
+**Version:** 1.0.0
+**Command:** `*qualify-task <task-name> [--batch]`
+**Execution Type:** Workflow Orchestrator
+**Model:** `Haiku` (orchestration only - workflow handles model selection)
+**Haiku Eligible:** YES — orchestration is deterministic
+**Orchestrator:** @pedro-valerio
+
+## Purpose
+
+Invocar `wf-model-tier-qualification.yaml` para testar empiricamente se uma task pode rodar em modelo mais barato (Haiku) mantendo qualidade.
+
+**Diferença do `*optimize`:**
+- `*optimize --hybrid` → Pipeline completo (análise + conversão + qualificação + GAP ZERO)
+- `*qualify-task` → APENAS qualificação empírica (Opus vs Haiku)
+
+## Usage
+
+```bash
+# Qualificar uma task específica
+*qualify-task pv-axioma-assessment
+
+# Qualificar com threshold customizado (default: 90%)
+*qualify-task validate-squad --threshold 0.85
+
+# Qualificar batch (todas tasks não-qualificadas)
+*qualify-task --batch
+
+# Qualificar batch de uma wave específica
+*qualify-task --batch --wave 1
+```
+
+## Veto Conditions
+
+- [ ] VETO se task não existe em `squads/squad-creator-pro/tasks/`
+- [ ] VETO se task não tem `test_input` no registry do workflow
+- [ ] VETO se já está qualificada em `model-routing.yaml` (use `--force` para re-testar)
+
+---
+
+## Workflow
+
+### Step 1: Pre-flight Validation
+
+```yaml
+preflight:
+  - check: "Task file exists?"
+    path: "squads/squad-creator-pro/tasks/{task_name}.md"
+    veto_if: "not exists"
+
+  - check: "Task in test_input_registry?"
+    lookup: "wf-model-tier-qualification.yaml → test_input_registry.{task_name}"
+    veto_if: "not found"
+    message: "Task needs test_input defined first. Add to workflow registry."
+
+  - check: "Already qualified?"
+    lookup: "config/model-routing.yaml → tasks.{task_name}.validated"
+    action_if_true: "Skip unless --force flag"
+```
+
+### Step 2: Load Workflow
+
+```yaml
+load_workflow:
+  file: "squads/squad-creator-pro/workflows/wf-model-tier-qualification.yaml"
+  action: "Read COMPLETE workflow definition"
+
+  extract:
+    - phases (4 phases + auto-improve loop)
+    - test_input_registry
+    - veto_conditions
+    - compensations
+```
+
+### Step 3: Execute Qualification
+
+```yaml
+execute:
+  # Invoke workflow with inputs
+  inputs:
+    task_name: "{from command}"
+    target_tier: "haiku"  # default
+    threshold: 0.90       # default, can override with --threshold
+
+  phases:
+    - phase_0: "PRE-FLIGHT"
+    - phase_1: "PARALLEL EXECUTION (Opus + Haiku)"
+    - phase_2: "COMPARE & AUTO-DECIDE"
+    - phase_3: "AUTO-IMPROVE LOOP (if needed)"
+    - phase_4: "GENERATE REPORT"
+
+  # Follow workflow phases exactly as defined
+  action: |
+    Execute wf-model-tier-qualification.yaml with the provided inputs.
+    The workflow handles:
+    - Running Opus baseline
+    - Running Haiku candidate
+    - Comparing results
+    - Applying compensations if needed
+    - Generating qualification report
+```
+
+### Step 4: Update Config
+
+```yaml
+update_config:
+  condition: "decision is QUALIFIED or QUALIFIED_WITH_COMPENSATION"
+
+  action: |
+    Update config/model-routing.yaml:
+
+    tasks:
+      {task_name}.md:
+        tier: haiku  # or sonnet if escalated
+        confidence: high
+        validated: true
+        test_date: "{today}"
+        test_result: "{summary from qualification report}"
+
+  script: |
+    node squads/squad-creator-pro/scripts/model-tier-validator.cjs \
+      update-routing {task_name} {recommended_tier} "{reason}"
+```
+
+### Step 5: Summary
+
+```yaml
+summary:
+  display: |
+    ╔══════════════════════════════════════════════════════════════════╗
+    ║  QUALIFICATION RESULT: {task_name}                               ║
+    ╠══════════════════════════════════════════════════════════════════╣
+    ║                                                                  ║
+    ║  Decision: {QUALIFIED | NEEDS_COMPENSATION | ESCALATE | OPUS}    ║
+    ║  Tier: {haiku | sonnet | opus}                                   ║
+    ║  Quality: {pct}% of Opus baseline                                ║
+    ║  Cost Savings: {savings_pct}%                                    ║
+    ║                                                                  ║
+    ║  Report: test-cases/{task_name}/qualification-report.yaml        ║
+    ╚══════════════════════════════════════════════════════════════════╝
+```
+
+---
+
+## Batch Mode
+
+```yaml
+batch_mode:
+  trigger: "--batch flag"
+
+  action: |
+    1. Read wf-model-tier-qualification.yaml → batch_mode.execution_order
+    2. Filter tasks not yet validated in model-routing.yaml
+    3. Execute qualification for each task sequentially
+    4. Generate consolidated QUALIFICATION-DASHBOARD.yaml
+
+  waves:
+    wave_1: "High-confidence Haiku candidates"
+    wave_2: "Admin tasks (script-based)"
+    wave_3: "Sonnet candidates tested with Haiku"
+
+  output:
+    - "test-cases/{task}/qualification-report.yaml" (per task)
+    - "test-cases/QUALIFICATION-DASHBOARD.yaml" (consolidated)
+```
+
+---
+
+## Examples
+
+### Example 1: Qualify Single Task
+
+```
+User: *qualify-task an-assess-sources
+
+Agent:
+## Step 1: Pre-flight ✅
+- Task file: squads/squad-creator-pro/tasks/an-assess-sources.md ✅
+- Test input: outputs/minds/pedro_valerio/sources/ ✅
+- Already qualified: NO
+
+## Step 2: Loading workflow...
+Read: wf-model-tier-qualification.yaml
+
+## Step 3: Executing qualification...
+[Phase 0: PRE-FLIGHT] ✅
+[Phase 1: PARALLEL EXECUTION]
+  - Opus baseline running...
+  - Haiku candidate running...
+[Phase 2: COMPARE]
+  - Opus: 8.5/10
+  - Haiku: 8.2/10 (96.5% quality)
+  - Decision: QUALIFIED ✅
+
+## Step 4: Updating config...
+Updated model-routing.yaml: an-assess-sources.md → haiku
+
+## Result
+╔══════════════════════════════════════════════════════════════════╗
+║  QUALIFICATION RESULT: an-assess-sources                         ║
+╠══════════════════════════════════════════════════════════════════╣
+║  Decision: QUALIFIED                                             ║
+║  Tier: haiku                                                     ║
+║  Quality: 96.5%                                                  ║
+║  Savings: 97%                                                    ║
+╚══════════════════════════════════════════════════════════════════╝
+```
+
+### Example 2: Batch Mode
+
+```
+User: *qualify-task --batch --wave 1
+
+Agent:
+## Batch Qualification: Wave 1 (10 tasks)
+
+| # | Task | Status | Tier | Quality | Savings |
+|---|------|--------|------|---------|---------|
+| 1 | pv-axioma-assessment | ✅ QUALIFIED | haiku | 95.5% | 97% |
+| 2 | validate-squad | ✅ QUALIFIED | haiku | 91% | 98% |
+| 3 | an-fidelity-score | ✅ SCRIPT | script | 100% | 100% |
+| 4 | an-diagnose-clone | ❌ OPUS REQUIRED | opus | 30% | 0% |
+...
+
+## Summary
+- Qualified: 8/10 (80%)
+- Script-only: 2/10 (20%)
+- Opus required: 1/10 (10%)
+- Total savings: ~85% token cost reduction
+
+Dashboard: test-cases/QUALIFICATION-DASHBOARD.yaml
+```
+
+---
+
+## Completion Criteria
+
+| Critério | Status |
+|----------|--------|
+| Pre-flight validation passed | [ ] |
+| Workflow executed completely | [ ] |
+| Qualification report generated | [ ] |
+| Config updated (if qualified) | [ ] |
+| Summary displayed | [ ] |
+
+---
+
+## Related
+
+- `wf-model-tier-qualification.yaml` - The actual workflow being invoked
+- `config/model-routing.yaml` - Where results are stored
+- `*optimize --hybrid` - Full pipeline (includes qualification)
+- `*lookup-model` - Read-only lookup (no qualification)
+- `*smoke-test-model-routing` - Test the routing system itself
+
+---
+
+_Task Version: 1.0.0_
+_Last Updated: 2026-02-12_
+
+*"Não assumir. Provar com dados."*

package/pro/squads/squad-creator-pro/tasks/reexecute-squad-phase.md

@@ -1,64 +1,64 @@
-# Task: Reexecute Squad Phase
-
-**Task ID:** reexecute-squad-phase
-**Version:** 1.0.0
-**Execution Type:** Interactive Hybrid
-**Purpose:** Re-run one workflow phase with backup and deterministic rollback path
-**Orchestrator:** @squad-chief
-**Mode:** Human-confirmed for destructive phase cleanup
-
-## Why
-
-```
-Brownfield improvements need safe retries.
-Reexecution must preserve recoverability.
-```
-
----
-
-## Inputs
-
-| Parameter | Type | Required | Description |
-|-----------|------|----------|-------------|
-| `squad_name` | string | Yes | Target squad |
-| `workflow_id` | string | Yes | Workflow to reexecute |
-| `phase_id` | string | Yes | Phase to rerun |
-| `reason` | text | Yes | Why rerun is required |
-
----
-
-## Safety Protocol
-
-1. Snapshot phase inputs/outputs.
-2. Commit backup with message: `backup: {squad_name} {workflow_id} {phase_id}`.
-3. Show impacted downstream phases.
-4. Require explicit confirmation.
-5. Clean only phase-scoped artifacts.
-6. Reexecute phase task list.
-7. Run targeted validation.
-
----
-
-## Output
-
-```yaml
-reexecution_report:
-  squad_name: "..."
-  workflow_id: "..."
-  phase_id: "..."
-  backup_ref: "git-commit-sha"
-  cleaned_artifacts:
-    - "..."
-  rerun_status: success | fail
-  downstream_impacts:
-    - "..."
-  rollback_instructions: "git checkout {sha} -- <paths>"
-```
-
----
-
-## Validation
-
-- Backup commit created before cleanup.
-- No non-phase artifacts removed.
-- Target phase completed with no blocking errors.
+# Task: Reexecute Squad Phase
+
+**Task ID:** reexecute-squad-phase
+**Version:** 1.0.0
+**Execution Type:** Interactive Hybrid
+**Purpose:** Re-run one workflow phase with backup and deterministic rollback path
+**Orchestrator:** @squad-chief
+**Mode:** Human-confirmed for destructive phase cleanup
+
+## Why
+
+```
+Brownfield improvements need safe retries.
+Reexecution must preserve recoverability.
+```
+
+---
+
+## Inputs
+
+| Parameter | Type | Required | Description |
+|-----------|------|----------|-------------|
+| `squad_name` | string | Yes | Target squad |
+| `workflow_id` | string | Yes | Workflow to reexecute |
+| `phase_id` | string | Yes | Phase to rerun |
+| `reason` | text | Yes | Why rerun is required |
+
+---
+
+## Safety Protocol
+
+1. Snapshot phase inputs/outputs.
+2. Commit backup with message: `backup: {squad_name} {workflow_id} {phase_id}`.
+3. Show impacted downstream phases.
+4. Require explicit confirmation.
+5. Clean only phase-scoped artifacts.
+6. Reexecute phase task list.
+7. Run targeted validation.
+
+---
+
+## Output
+
+```yaml
+reexecution_report:
+  squad_name: "..."
+  workflow_id: "..."
+  phase_id: "..."
+  backup_ref: "git-commit-sha"
+  cleaned_artifacts:
+    - "..."
+  rerun_status: success | fail
+  downstream_impacts:
+    - "..."
+  rollback_instructions: "git checkout {sha} -- <paths>"
+```
+
+---
+
+## Validation
+
+- Backup commit created before cleanup.
+- No non-phase artifacts removed.
+- Target phase completed with no blocking errors.