aiox-core 5.0.2 → 5.0.3

package/pro/squads/squad-creator-pro/tasks/extract-sop.md

@@ -0,0 +1,320 @@
+---
+task: Extract SOP from Transcript
+responsavel: "@sop-extractor"
+responsavel_type: agent
+atomic_layer: task
+elicit: true
+phase: discovery
+
+# Dependencies
+templates:
+  - pop-extractor-prompt
+config:
+  - squad-config
+---
+
+# Extract SOP from Transcript
+
+**Squad:** squad-creator
+**Phase:** Discovery
+**Agent:** @sop-extractor
+**Pattern:** SC-PE-001 (SOP Extraction Standard)
+
+## Purpose
+
+Extract a complete, AIOS-ready Standard Operating Procedure (SOP) from a meeting transcript where someone explained a business process. The output is structured to enable immediate squad creation for hybrid automation.
+
+## Task Anatomy (HO-TP-001)
+
+| Field | Value |
+|-------|-------|
+| task_name | Extract SOP from Transcript |
+| status | pending |
+| responsible_executor | @sop-extractor |
+| execution_type | Hybrid (Agent extracts, Human validates) |
+| estimated_time | 1-2h per process |
+| input | transcript, domain_context |
+| output | sop_document, squad_blueprint, gap_report |
+| action_items | See Execution section |
+| acceptance_criteria | All 11 parts completed, gaps documented |
+
+## Inputs
+
+| Input | Type | Required | Description |
+|-------|------|----------|-------------|
+| transcript | text/file | Yes | Meeting transcript (text, audio transcription, or video transcript) |
+| domain_context | string | No | Business domain/area for terminology context |
+| existing_docs | file[] | No | Current SOPs, manuals, or process docs |
+| process_owner | string | No | Who to validate extracted SOP with |
+
+## Outputs
+
+| Output | Type | Description |
+|--------|------|-------------|
+| sop_document | MD | Complete SOP following SC-PE-001 template |
+| squad_blueprint | YAML | Ready-to-use squad structure (Part 8) |
+| gap_report | MD | Missing information and clarifying questions |
+| automation_analysis | MD | Summary of automation potential (Part 7) |
+
+## Execution
+
+### Step 0: Fetch Transcript (Data Source Resolution)
+
+1. Load configuration from `config/squad-config.yaml`
+2. Read `data_sources.transcripts.active_source`
+3. Fetch transcript based on source type:
+
+**Source Resolution:**
+
+```yaml
+# Read active_source from squad-config.yaml
+active_source: {{squad_config.data_sources.transcripts.active_source}}
+
+# Execute appropriate fetch
+if active_source == "supabase":
+  - Connect using env vars from squad_config.data_sources.transcripts.sources.supabase.connection
+  - Execute query from squad_config.data_sources.transcripts.sources.supabase.query
+  - Map fields using field_mapping
+
+elif active_source == "local_file":
+  - Read from squad_config.data_sources.transcripts.sources.local_file.config.base_path
+  - Parse based on file extension
+  - Map fields if JSON
+
+elif active_source == "api":
+  - Call endpoint from squad_config.data_sources.transcripts.sources.api.endpoints.get
+  - Parse response using data_path
+  - Map fields
+
+elif active_source == "direct":
+  - Use transcript passed as input parameter
+```
+
+**Output:** Transcript object with standard schema:
+```yaml
+transcript:
+  transcript_id: string
+  transcript_content: string
+  transcript_source: string (optional)
+  transcript_url: string (optional)
+  transcript_duration: number (optional)
+  transcript_participants: array (optional)
+  transcript_date: date (optional)
+  transcript_language: string (optional)
+  transcript_metadata: object (optional)
+```
+
+**Elicit if source not configured:**
+```
+Which transcript source should I use?
+1. Supabase database (default)
+2. Local file (provide path)
+3. External API (requires config)
+4. Direct input (paste transcript)
+```
+
+---
+
+### Step 1: Transcript Preparation
+
+1. Validate transcript was fetched successfully
+2. Identify language and normalize if needed
+3. Mark timestamps or speaker changes
+4. Identify main process explainer
+5. Apply chunking if transcript exceeds `max_tokens_per_chunk`
+
+**Elicit if unclear:**
+```
+What is the business domain of this process?
+Who is the process owner to validate with?
+Are there existing documents to cross-reference?
+```
+
+### Step 2: First Pass - Structure Identification
+
+1. Identify process name and objective
+2. List all mentioned steps (even informal)
+3. Note all people/roles mentioned
+4. Capture all tools/systems referenced
+5. Mark decision points ("if", "when", "depends")
+
+**Output:** Raw extraction notes
+
+### Step 3: Second Pass - Task Anatomy Mapping
+
+For each identified step, extract:
+
+```yaml
+step_template:
+  task_name: "[Verb] + [Object]"
+  responsible_executor: "[Role or @agent]"
+  execution_type: "Human | Agent | Hybrid | Worker"
+  estimated_time: "[Xh/Xm]"
+  input: ["list of inputs"]
+  output: ["list of outputs"]
+  action_items: ["atomic steps"]
+  acceptance_criteria: ["how to verify success"]
+```
+
+**Classification Guide:**
+
+| Cognitive Signal | Executor Type |
+|------------------|---------------|
+| "I look at...", "I check..." | Agent (perception) |
+| "I decide based on...", "It depends..." | Hybrid (judgment) |
+| "I talk to...", "I convince..." | Human (relationship) |
+| "I copy...", "I move...", "I send..." | Worker (deterministic) |
+
+### Step 4: Third Pass - Decision Rules Extraction
+
+For each "depends", "usually", "sometimes":
+
+1. Identify the condition
+2. Identify possible outcomes
+3. Translate to IF/THEN rule
+4. Mark if automatable or requires human judgment
+
+**Output:** Decision rules table + heuristics list
+
+### Step 5: Automation Analysis (PV_PM_001)
+
+For each step, evaluate:
+
+| Criterion | Question |
+|-----------|----------|
+| Frequency | How often? (>4x/mo = high) |
+| Impact | What if it fails? (business impact) |
+| Automatability | Can code/AI do it? (determinism level) |
+| Guardrails | Can we add safeguards? (required for automation) |
+
+**Apply Decision Matrix:**
+- AUTOMATE: High freq + High impact + High auto + Has guardrails
+- DELEGATE: High freq + High impact + Low auto
+- KEEP_MANUAL: Low freq + High impact
+- ELIMINATE: Low freq + Low impact
+- VETO: No guardrails possible
+
+### Step 6: Quality Assessment (META-AXIOMAS)
+
+Score the process on 10 dimensions (0-10):
+
+1. Truthfulness - Is the process described accurately?
+2. Coherence - Do steps align logically?
+3. Strategic Alignment - Does it serve business goals?
+4. Operational Excellence - Is it efficient?
+5. Innovation Capacity - Can it evolve?
+6. Risk Management - Are risks addressed?
+7. Resource Optimization - Is it lean?
+8. Stakeholder Value - Does it serve users?
+9. Sustainability - Is it maintainable?
+10. Adaptability - Can it handle change?
+
+**Threshold:** Overall ≥7.0 to proceed
+
+### Step 7: Squad Blueprint Generation
+
+Based on extracted data, generate:
+
+1. **Agents needed** - One per major responsibility
+2. **Tasks to create** - One per workflow phase
+3. **Checkpoints** - Where human validation required
+4. **Guardrails** - Required safeguards per task
+5. **Workflow YAML** - Orchestration definition
+
+### Step 8: Gap Analysis
+
+Document all:
+
+1. **Missing information** - What wasn't explained
+2. **Ambiguities** - Multiple interpretations possible
+3. **Assumptions** - What was inferred (mark [INFERRED])
+4. **Red flags** - Single points of failure, undocumented exceptions
+
+**Output:** Gap report with clarifying questions
+
+### Step 9: Document Assembly
+
+Assemble final SOP using template `pop-extractor-prompt.md`:
+
+1. Fill all 11 parts
+2. Generate Mermaid diagrams
+3. Complete all tables
+4. Add glossary terms
+5. Set metadata
+
+## Validation
+
+**Validation Type:** Human review required
+
+**Validation Checklist:**
+
+- [ ] All 11 parts of SC-PE-001 completed
+- [ ] Task Anatomy (8 fields) for each step
+- [ ] Executor type assigned to each step
+- [ ] Decision rules extracted as IF/THEN
+- [ ] Automation analysis completed (PV_PM_001)
+- [ ] META-AXIOMAS score ≥7.0
+- [ ] Squad blueprint generated
+- [ ] Gaps documented with questions
+- [ ] No [INFERRED] items left unvalidated
+
+**Quality Gate:**
+
+| Criterion | Threshold | Action if Failed |
+|-----------|-----------|------------------|
+| Parts completed | 11/11 | Block until complete |
+| Steps with Task Anatomy | 100% | Block until complete |
+| Gaps documented | All listed | Review with process owner |
+| META-AXIOMAS score | ≥7.0 | Review weak dimensions |
+
+## Handoff
+
+```yaml
+handoff:
+  to: "create-squad"
+  trigger: "sop_validated = true"
+  data_transferred:
+    - sop_document
+    - squad_blueprint
+    - gap_report
+  validation: "Process owner confirms accuracy"
+```
+
+## Tools
+
+| Tool | Purpose |
+|------|---------|
+| Transcript parser | Extract text from audio/video |
+| Mermaid | Generate flow diagrams |
+| YAML validator | Validate squad blueprint |
+
+## Templates
+
+- `pop-extractor-prompt.md` - SOP extraction template (SC-PE-001)
+
+## Error Handling
+
+| Error | Cause | Resolution |
+|-------|-------|------------|
+| Incomplete transcript | Audio quality, speaker overlap | Request clarification from process owner |
+| Conflicting information | Multiple speakers disagree | Document both versions, flag for validation |
+| Missing steps | Tacit knowledge not verbalized | Add to gaps, schedule follow-up interview |
+| Unclear executor | Role not specified | Default to Hybrid, flag for validation |
+
+## Examples
+
+**Good transcript signals:**
+- "First, I do X, then Y, then Z" → Clear sequence
+- "If the client says no, I do A, otherwise B" → Decision rule
+- "I always check this before proceeding" → Precondition
+- "This takes about 30 minutes" → Time estimate
+
+**Red flag signals:**
+- "It depends" (without criteria) → Needs clarification
+- "João knows how to do this" → Single point of failure
+- "We figure it out" → Undocumented exception
+- "It's always been this way" → May be obsolete
+
+---
+
+**Pattern Compliance:** SC-PE-001 ✓ | HO-TP-001 ✓ | HO-TP-002 ✓