aiox-core 5.0.4 → 5.0.7

package/pro/squads/squad-creator-pro/workflows/wf-discover-tools.yaml

@@ -1,124 +1,334 @@
-# wf-discover-tools.yaml
-# Workflow: Deep Tool Discovery with Parallel Sub-Agents
-# Version: 2.0
-# Last Updated: 2026-02-03
-
-# ═══════════════════════════════════════════════════════════════════════════════
-# WORKFLOW METADATA
-# ═══════════════════════════════════════════════════════════════════════════════
-
 workflow:
   id: wf-discover-tools
-  name: "Deep Tool Discovery"
-  version: "2.0"
+  name: Deep Tool Discovery
+  version: "3.2"
   purpose: |
     Orquestrar pesquisa PROFUNDA e PARALELA de ferramentas que potencializam
     os entregáveis de um squad, usando subagentes especializados para cada
     categoria de ferramenta.
   orchestrator: "@squad-chief"
-  mode: "autonomous"
-
+  mode: autonomous
+  description: Legacy phase-based workflow for deep tool discovery.
+  type: discovery
+  state_machine:
+    states: [init, domain-context-validation, internal-infrastructure-discovery, capability-gap-analysis, parallel-deep-search, comparative-evaluation, decision-matrix, handoff, complete, failed]
+    transitions:
+      - trigger: start
+        source: init
+        dest: domain-context-validation
+      - trigger: domain_context_validation_complete
+        source: domain-context-validation
+        dest: internal-infrastructure-discovery
+      - trigger: domain_context_validation_failed
+        source: domain-context-validation
+        dest: failed
+      - trigger: internal_infrastructure_discovery_complete
+        source: internal-infrastructure-discovery
+        dest: capability-gap-analysis
+      - trigger: internal_infrastructure_discovery_failed
+        source: internal-infrastructure-discovery
+        dest: failed
+      - trigger: capability_gap_analysis_complete
+        source: capability-gap-analysis
+        dest: parallel-deep-search
+      - trigger: capability_gap_analysis_failed
+        source: capability-gap-analysis
+        dest: failed
+      - trigger: parallel_deep_search_complete
+        source: parallel-deep-search
+        dest: comparative-evaluation
+      - trigger: parallel_deep_search_failed
+        source: parallel-deep-search
+        dest: failed
+      - trigger: comparative_evaluation_complete
+        source: comparative-evaluation
+        dest: decision-matrix
+      - trigger: comparative_evaluation_failed
+        source: comparative-evaluation
+        dest: failed
+      - trigger: decision_matrix_complete
+        source: decision-matrix
+        dest: handoff
+      - trigger: decision_matrix_failed
+        source: decision-matrix
+        dest: failed
+      - trigger: handoff_complete
+        source: handoff
+        dest: complete
+      - trigger: handoff_failed
+        source: handoff
+        dest: failed
+      - trigger: fail
+        source: "*"
+        dest: failed
+  sequence:
+    - step: phase-minus-2
+      id: phase-minus-2
+      phase: 1
+      phase_name: DOMAIN CONTEXT VALIDATION
+      agent: "@squad-chief"
+      action: execute_domain-context-validation
+      task_ref: discover-tools-execute
+      next: phase-minus-1
+    - step: phase-minus-1
+      id: phase-minus-1
+      phase: 2
+      phase_name: INTERNAL INFRASTRUCTURE DISCOVERY
+      agent: "@squad-chief"
+      action: execute_internal-infrastructure-discovery
+      task_ref: discover-tools-execute
+      next: phase-0
+    - step: phase-0
+      id: phase-0
+      phase: 3
+      phase_name: CAPABILITY GAP ANALYSIS
+      agent: "@squad-chief"
+      action: execute_capability-gap-analysis
+      task_ref: discover-tools-execute
+      next: phase-1
+    - step: phase-1
+      id: phase-1
+      phase: 4
+      phase_name: PARALLEL DEEP SEARCH
+      agent: "@squad-chief"
+      action: execute_this-phase-launches-6-subagents-simultaneously-using-the-task-to
+      task_ref: discover-tools-execute
+      next: phase-2
+    - step: phase-2
+      id: phase-2
+      phase: 5
+      phase_name: COMPARATIVE EVALUATION
+      agent: "@squad-chief"
+      action: execute_comparative-evaluation
+      task_ref: discover-tools-execute
+      next: phase-3
+    - step: phase-3
+      id: phase-3
+      phase: 6
+      phase_name: DECISION MATRIX & INTEGRATION PLAN
+      agent: "@squad-chief"
+      action: execute_decision-matrix-integration-plan
+      task_ref: discover-tools-execute
+      next: phase-4
+    - step: phase-4
+      id: phase-4
+      phase: 7
+      phase_name: HANDOFF
+      agent: "@squad-chief"
+      action: execute_handoff
+      task_ref: discover-tools-execute
+    - workflow_end:
+        id: complete
+        action: workflow_complete
   philosophy: |
+    "INTERNAL-FIRST: use what AIOX already has before searching outside."
+
     "A squad should leverage ALL available tools to deliver maximum value
     with minimum user intervention."
 
     Este workflow NÃO é um check superficial - é uma pesquisa PROFUNDA
-    que lança 5 subagentes em PARALELO, cada um especializado em uma
+    que lança 6 subagentes em PARALELO, cada um especializado em uma
     categoria de ferramenta.
-
-# ═══════════════════════════════════════════════════════════════════════════════
-# EXECUTION STRATEGY
-# ═══════════════════════════════════════════════════════════════════════════════
-
 execution_strategy:
-  mode: "parallel_then_synthesize"
+  mode: parallel_then_synthesize
   description: |
-    1. PHASE 0: Capability Gap Analysis (sequential, fast)
-    2. PHASE 1: Deep Search (5 subagents in PARALLEL)
+    1. PHASE -2: Domain Context Validation (sequential, mandatory)
+    2. PHASE -1: Internal Infrastructure Discovery (sequential, mandatory)
+    3. PHASE 0: Capability Gap Analysis + Local Skills/Squads Scan (sequential, fast)
+    4. PHASE 1: Deep Search (6 subagents in PARALLEL, external-only gaps)
        - MCP Discovery Agent
        - API Discovery Agent
        - CLI Discovery Agent
        - Library Discovery Agent
        - GitHub Project Discovery Agent
-    3. PHASE 2: Synthesis & Scoring (sequential, after all agents complete)
-    4. PHASE 3: Decision Matrix & Integration Plan
-
+       - Skill Discovery Agent
+    5. PHASE 2: Synthesis & Scoring (sequential, after all agents complete)
+    6. PHASE 3: Decision Matrix & Integration Plan
   parallelization:
     enabled: true
-    max_concurrent_agents: 5
-    timeout_per_agent: "15 minutes"
-    fail_strategy: "continue_others"  # Don't block if one agent fails
-
-# ═══════════════════════════════════════════════════════════════════════════════
-# INPUTS
-# ═══════════════════════════════════════════════════════════════════════════════
-
+    max_concurrent_agents: 6
+    timeout_per_agent: 15 minutes
+    fail_strategy: continue_others
 inputs:
   required:
     - name: domain
       type: string
-      description: "Squad domain (e.g., 'copywriting', 'legal', 'cybersecurity')"
-      example: "copywriting"
-
+      description: Squad domain (e.g., 'copywriting', 'legal', 'cybersecurity')
+      example: copywriting
     - name: use_cases
       type: list
-      description: "Key use cases the squad must handle"
-      example: ["sales pages", "email sequences", "ad copy"]
-
+      description: Key use cases the squad must handle
+      example:
+        - sales pages
+        - email sequences
+        - ad copy
   optional:
     - name: existing_tools
       type: list
-      description: "Tools already in use (to avoid duplicates)"
+      description: Tools already in use (to avoid duplicates)
       default: []
-
     - name: budget_tier
       type: enum
-      values: ["free_only", "low_cost", "enterprise"]
-      default: "low_cost"
-
+      values:
+        - free_only
+        - low_cost
+        - enterprise
+      default: low_cost
     - name: priority_capabilities
       type: list
-      description: "Capabilities to prioritize in search"
+      description: Capabilities to prioritize in search
       default: []
-
     - name: search_depth
       type: enum
-      values: ["quick", "standard", "exhaustive"]
-      default: "standard"
+      values:
+        - quick
+        - standard
+        - exhaustive
+      default: standard
       description: |
         quick: 5 results per category, 5 min timeout
         standard: 15 results per category, 10 min timeout
         exhaustive: 30 results per category, 20 min timeout
-
-# ═══════════════════════════════════════════════════════════════════════════════
-# PRECONDITIONS
-# ═══════════════════════════════════════════════════════════════════════════════
-
 preconditions:
-  - "squad-chief agent is active"
-  - "WebSearch/EXA tool available"
-  - "WebFetch tool available"
-  - "Task tool available (for spawning subagents)"
-  - "Write permissions for data/tool-registry.yaml"
+  - squad-chief agent is active
+  - domain scope validated from canonical artifacts (no name-only inference)
+  - data/internal-infrastructure-library.yaml exists
+  - WebSearch/EXA tool available
+  - WebFetch tool available
+  - Task tool available (for spawning subagents)
+  - Write permissions for .aiox/squad-runtime/discovery/{domain}/
+phases:
+  - id: phase_minus_2
+    name: DOMAIN CONTEXT VALIDATION
+    task_ref: discover-tools-execute
+    purpose: Resolve real domain scope from canonical artifacts before any discovery
+    duration: 2-6 minutes
+    mode: sequential
+    steps:
+      - id: step_m2_1
+        name: Resolve Canonical Scope
+        action: Load canonical files for the requested domain/squad and extract explicit purpose
+        logic: |
+          canonical_sources_priority:
+            - squads/{domain}/squad.yaml
+            - squads/{domain}/config.yaml
+            - squads/{domain}/README.md
+            - {registry_path}
+            - related tasks/workflows/agents
+
+          rules:
+            - never infer scope from domain name/slug alone
+            - extract explicit domain + purpose + use_cases statements
+            - mark missing/conflicting evidence
+      - id: step_m2_2
+        name: Build Evidence Pack
+        action: Create canonical evidence with status and contradiction log
+        logic: |
+          canonical_scope:
+            declared_domain: ""
+            declared_purpose: ""
+            evidence:
+              - path: ""
+                status: "implemented|partial|concept"
+                extracted_scope: ""
+            contradictions: []
+            confidence: "high|medium|low"
+      - id: step_m2_3
+        name: Apply Ambiguity Gate
+        action: Block discovery when scope is unresolved or inferred by name
+        logic: |
+          fail_when:
+            - no explicit purpose in canonical files
+            - conflicting canonical sources
+            - confidence == low
+            - scope came from name-only inference
+
+          on_fail:
+            - halt external discovery
+            - request user clarification with explicit options
+        output:
+          scope_validation:
+            status: verified|partial|unverified
+            requires_clarification: true/false
+            evidence_paths: []
+            contradictions: []
+    checkpoint:
+      id: CP_DOMAIN_INTENT
+      name: Domain Context Validation Gate
+      blocking: true
+      criteria:
+        - scope_validation_status: verified
+        - name_only_inference_used: false
+        - contradictions_resolved_or_logged: true
+  - id: phase_minus_1
+    name: INTERNAL INFRASTRUCTURE DISCOVERY
+    task_ref: discover-tools-execute
+    purpose: Map internal capabilities first and constrain external search scope
+    duration: 3-8 minutes
+    mode: sequential
+    depends_on:
+      - phase_minus_2
+    steps:
+      - id: step_m1_1
+        name: Load Internal Library
+        action: Read data/internal-infrastructure-library.yaml
+        logic: |
+          load: "data/internal-infrastructure-library.yaml"
+          assert_exists:
+            - internal_assets
+            - decision_policy
+            - output_contract
+      - id: step_m1_2
+        name: Map Internal Coverage
+        action: Match required capabilities against internal assets
+        logic: |
+          for each use_case in inputs.use_cases:
+            derive required_capabilities
 
-# ═══════════════════════════════════════════════════════════════════════════════
-# PHASES
-# ═══════════════════════════════════════════════════════════════════════════════
+          for each capability in required_capabilities:
+            match against internal_assets[].provides[].capability
 
-phases:
-  # ─────────────────────────────────────────────────────────────────────────────
-  # PHASE 0: CAPABILITY GAP ANALYSIS
-  # ─────────────────────────────────────────────────────────────────────────────
+            classify:
+              - INTERNAL_COVERED
+              - INTERNAL_PARTIAL
+              - INTERNAL_GAP
+        output:
+          internal_coverage:
+            covered: []
+            partial: []
+            gaps: []
+      - id: step_m1_3
+        name: Define External-Only Scope
+        action: Prepare external search list based on true internal gaps
+        logic: |
+          external_search_scope.required = internal_coverage.gaps
+          external_search_scope.optional = internal_coverage.partial
+          external_search_scope.excluded = internal_coverage.covered
+        output:
+          external_search_scope:
+            required: []
+            optional: []
+            excluded: []
+    checkpoint:
+      id: CP_INTERNAL_FIRST
+      name: Internal Coverage Gate
+      blocking: true
+      criteria:
+        - internal_library_loaded: true
+        - internal_coverage_mapped: true
+        - external_scope_defined: true
   - id: phase_0
-    name: "CAPABILITY GAP ANALYSIS"
-    purpose: "Map required capabilities and identify gaps that need tools"
-    duration: "5-10 minutes"
-    mode: "sequential"
-
+    name: CAPABILITY GAP ANALYSIS
+    task_ref: discover-tools-execute
+    purpose: Map required capabilities and identify gaps that need tools
+    duration: 5-10 minutes
+    mode: sequential
+    depends_on:
+      - phase_minus_1
     steps:
       - id: step_0_1
-        name: "Map Use Case Requirements"
-        action: "Analyze each use case to identify required capabilities"
+        name: Map Use Case Requirements
+        action: Analyze each use case to identify required capabilities
         logic: |
           for each use_case in inputs.use_cases:
             analyze:
@@ -127,19 +337,21 @@ phases:
               - OUTPUT_NEEDS: What should be produced?
               - INTEGRATION_NEEDS: What external systems?
               - AUTOMATION_POTENTIAL: What can be automated?
-
         output:
           capability_requirements:
-            - use_case: "sales pages"
+            - use_case: sales pages
               capabilities:
-                - { name: "competitor_analysis", priority: "high" }
-                - { name: "headline_generation", priority: "high" }
-                - { name: "screenshot_capture", priority: "medium" }
-
+                - name: competitor_analysis
+                  priority: high
+                - name: headline_generation
+                  priority: high
+                - name: screenshot_capture
+                  priority: medium
       - id: step_0_2
-        name: "Check Existing Coverage"
-        action: "Load tool-registry.yaml and check what's already covered"
+        name: Check Existing Coverage
+        action: Load tool-registry.yaml and check what's already covered
         logic: |
+          load: "data/internal-infrastructure-library.yaml"
           load: "data/tool-registry.yaml"
 
           for each capability in capability_requirements:
@@ -149,72 +361,132 @@ phases:
               - known_mcp_coverage
 
             classify:
+              - INTERNAL_COVERED: Internal tool exists and is ready
+              - INTERNAL_PARTIAL: Internal tool exists but with limitations
               - COVERED: Tool exists and is installed
               - KNOWN: Tool exists but not installed
               - GAP: No known tool for this capability
-
         output:
           coverage_report:
             covered: []
             known_not_installed: []
-            gaps: []  # These go to deep search
-
+            gaps: []
+      - id: step_0_2b
+        name: Scan Local Skills & Squads
+        action: Scan existing skills and squad capabilities for domain coverage
+        logic: |
+          # 1. Scan project-level skills
+          scan: ".claude/skills/*/SKILL.md"
+          for each skill:
+            extract:
+              - name (from frontmatter)
+              - description (from frontmatter)
+              - capabilities (inferred from description + body)
+              - agent (if skill delegates to specific agent)
+              - tools_used (allowed-tools from frontmatter)
+
+            match_against: capability_requirements
+            classify:
+              - COVERS: Skill directly solves this capability
+              - ADAPTABLE: Skill partially covers, can be forked/extended
+              - UNRELATED: No match
+
+          # 2. Scan squad capabilities
+          scan: "squads/*/config.yaml"
+          for each squad:
+            extract:
+              - domain
+              - agents (names + specialties)
+              - tasks (names + purposes)
+              - workflows (names + purposes)
+              - skills (if any embedded)
+
+            match_against: capability_requirements
+            classify:
+              - REUSABLE: Squad component directly applicable
+              - ADAPTABLE: Component can be modified for this domain
+              - UNRELATED: Different domain, no overlap
+
+          # 3. Scan global skills
+          scan: "~/.claude/skills/*/SKILL.md"
+          for each skill:
+            same extraction and matching as project skills
+        output:
+          local_skills_coverage:
+            skills_scanned: "N"
+            squads_scanned: "N"
+            covers:
+              - capability: web_research
+                covered_by:
+                  - type: skill
+                    name: tech-search
+                    path: .claude/skills/tech-search/SKILL.md
+            adaptable:
+              - capability: competitor_analysis
+                adaptable_from:
+                  - type: squad
+                    name: spy
+                    component: deep-research task
+                    effort: low
+            gaps: []
       - id: step_0_3
-        name: "Prioritize Gaps"
-        action: "Score and prioritize capability gaps for deep search"
+        name: Prioritize Gaps
+        action: Score and prioritize capability gaps for deep search
+        note: Only prioritize INTERNAL_GAP and justified INTERNAL_PARTIAL capabilities
         scoring:
           impact:
             weight: 0.35
-            question: "How much value does filling this gap add?"
+            question: How much value does filling this gap add?
           frequency:
             weight: 0.25
-            question: "How often is this capability needed?"
+            question: How often is this capability needed?
           user_dependency:
             weight: 0.25
-            question: "How much manual work does the gap require?"
+            question: How much manual work does the gap require?
           uniqueness:
             weight: 0.15
-            question: "Is this a differentiating capability?"
-
+            question: Is this a differentiating capability?
         output:
           prioritized_gaps:
-            - capability: "competitor_analysis"
+            - capability: competitor_analysis
               priority_score: 8.7
-              search_keywords: ["competitor", "analysis", "monitoring", "scraping"]
-            - capability: "email_automation"
+              search_keywords:
+                - competitor
+                - analysis
+                - monitoring
+                - scraping
+            - capability: email_automation
               priority_score: 8.2
-              search_keywords: ["email", "smtp", "sendgrid", "mailgun"]
-
+              search_keywords:
+                - email
+                - smtp
+                - sendgrid
+                - mailgun
     checkpoint:
       id: CP_GAP_ANALYSIS
-      name: "Gap Analysis Complete"
+      name: Gap Analysis Complete
       blocking: true
       criteria:
         - capabilities_mapped: true
+        - local_skills_scanned: true
+        - local_squads_scanned: true
         - gaps_identified: ">= 1"
         - gaps_prioritized: true
-
-  # ─────────────────────────────────────────────────────────────────────────────
-  # PHASE 1: PARALLEL DEEP SEARCH
-  # ─────────────────────────────────────────────────────────────────────────────
+        - skills_coverage_report_generated: true
   - id: phase_1
-    name: "PARALLEL DEEP SEARCH"
-    purpose: "Launch 5 specialized subagents to search each tool category"
-    duration: "10-20 minutes (parallel)"
-    mode: "parallel"
-
+    name: PARALLEL DEEP SEARCH
+    task_ref: discover-tools-execute
+    purpose: Launch 6 specialized subagents to search each tool category for external-only gaps
+    duration: 10-20 minutes (parallel)
+    mode: parallel
     description: |
-      This phase launches 5 subagents SIMULTANEOUSLY using the Task tool.
+      This phase launches 6 subagents SIMULTANEOUSLY using the Task tool.
       Each subagent is specialized in searching one category of tools.
-      All agents receive the same inputs (domain, gaps, keywords) but
+      All agents receive the same inputs (domain, external-only gaps, keywords) but
       search in different ecosystems.
-
     parallel_agents:
-      # ─────────────────────────────────────────────────────────────────────────
-      # AGENT 1: MCP DISCOVERY
-      # ─────────────────────────────────────────────────────────────────────────
       - id: agent_mcp
-        name: "MCP Discovery Agent"
+        name: MCP Discovery Agent
         task_prompt: |
           You are a specialized MCP (Model Context Protocol) discovery agent.
 
@@ -225,9 +497,9 @@ phases:
           ## Search Strategy
 
           ### 1. Official Sources (Priority 1)
-          - https://github.com/modelcontextprotocol/servers
-          - https://github.com/anthropics/anthropic-tools
-          - https://glama.ai/mcp/servers
+          - GitHub MCP server registry
+          - Anthropic tools registry
+          - Glama MCP directory
 
           ### 2. Community Sources (Priority 2)
           Search GitHub with these queries:
@@ -264,16 +536,11 @@ phases:
           ## Output Format
           Return a YAML list of all MCPs found, sorted by relevance to our gaps.
           Include at least {search_depth_count} results.
-
-        subagent_type: "Explore"
+        subagent_type: Explore
         timeout: "{search_timeout}"
-        output_key: "mcp_results"
-
-      # ─────────────────────────────────────────────────────────────────────────
-      # AGENT 2: API DISCOVERY
-      # ─────────────────────────────────────────────────────────────────────────
+        output_key: mcp_results
       - id: agent_api
-        name: "API Discovery Agent"
+        name: API Discovery Agent
         task_prompt: |
           You are a specialized API discovery agent.
 
@@ -284,9 +551,9 @@ phases:
           ## Search Strategy
 
           ### 1. API Directories
-          - RapidAPI: https://rapidapi.com/search/{domain}
-          - Public APIs: https://github.com/public-apis/public-apis
-          - API List: https://apilist.fun
+          - RapidAPI directory search
+          - Public APIs directory
+          - API List directory
           - ProgrammableWeb categories
 
           ### 2. Search Queries
@@ -354,16 +621,11 @@ phases:
           3. Documentation quality
 
           Include at least {search_depth_count} results.
-
-        subagent_type: "Explore"
+        subagent_type: Explore
         timeout: "{search_timeout}"
-        output_key: "api_results"
-
-      # ─────────────────────────────────────────────────────────────────────────
-      # AGENT 3: CLI DISCOVERY
-      # ─────────────────────────────────────────────────────────────────────────
+        output_key: api_results
       - id: agent_cli
-        name: "CLI Tool Discovery Agent"
+        name: CLI Tool Discovery Agent
         task_prompt: |
           You are a specialized CLI tool discovery agent.
 
@@ -428,16 +690,11 @@ phases:
           3. Cross-platform support
 
           Include at least {search_depth_count} results.
-
-        subagent_type: "Explore"
+        subagent_type: Explore
         timeout: "{search_timeout}"
-        output_key: "cli_results"
-
-      # ─────────────────────────────────────────────────────────────────────────
-      # AGENT 4: LIBRARY DISCOVERY
-      # ─────────────────────────────────────────────────────────────────────────
+        output_key: cli_results
       - id: agent_library
-        name: "Library Discovery Agent"
+        name: Library Discovery Agent
         task_prompt: |
           You are a specialized library/SDK discovery agent.
 
@@ -509,16 +766,11 @@ phases:
           3. Quality metrics
 
           Include at least {search_depth_count} results.
-
-        subagent_type: "Explore"
+        subagent_type: Explore
         timeout: "{search_timeout}"
-        output_key: "library_results"
-
-      # ─────────────────────────────────────────────────────────────────────────
-      # AGENT 5: GITHUB PROJECT DISCOVERY
-      # ─────────────────────────────────────────────────────────────────────────
+        output_key: library_results
       - id: agent_github
-        name: "GitHub Project Discovery Agent"
+        name: GitHub Project Discovery Agent
         task_prompt: |
           You are a specialized GitHub project discovery agent.
 
@@ -598,12 +850,96 @@ phases:
           3. Relevance to our gaps
 
           Include at least {search_depth_count} results.
-
-        subagent_type: "Explore"
+        subagent_type: Explore
         timeout: "{search_timeout}"
-        output_key: "github_results"
+        output_key: github_results
+      - id: agent_skill
+        name: Skill Discovery Agent
+        task_prompt: |
+          You are a specialized AI Skill & Prompt discovery agent.
+
+          ## Your Mission
+          Find reusable AI skills, prompt templates, agent frameworks, and workflow
+          patterns that can fill these capability gaps for a {domain} squad:
+          {prioritized_gaps}
+
+          IMPORTANT: The local skills scan (Phase 0) already identified these
+          LOCAL coverages: {local_skills_coverage}
+          Focus your search on EXTERNAL/COMMUNITY sources that ADD to local coverage.
+
+          ## Search Strategy
+
+          ### 1. Claude Code / Anthropic Ecosystem
+          - GitHub: "claude code skills" OR "claude skills" OR ".claude/skills"
+          - GitHub: "claude agent" OR "claude code agent" OR ".claude/agents"
+          - GitHub topic: "claude-code", "anthropic-tools", "claude-agent"
+
+          ### 2. AI Agent Framework Skills
+          - LangChain Hub
+          - CrewAI
+          - AutoGen
+          - Semantic Kernel
+          - GitHub: "awesome ai agents {domain}"
+          - GitHub: "ai agent {domain} prompt"
+
+          ### 3. Prompt Libraries & Templates
+          - Prompt marketplaces and prompt directories
+          - Community prompt hubs
+          - Domain-specific prompt libraries
+          - GitHub: "awesome prompts {domain}"
+          - GitHub: "system prompt {domain}"
+
+          ### 4. Workflow & Automation Patterns
+          - GitHub: "{domain} automation workflow"
+          - GitHub: "ai workflow {domain}"
+          - n8n workflow templates
+
+          ### 5. Filter Criteria
+          - Has clear documentation or README
+          - Reusable (not one-off personal projects)
+          - Domain-relevant or easily adaptable
+          - Active maintenance preferred but not required
+          - License allows adaptation (MIT, Apache, CC-BY preferred)
+
+          ## For Each Skill/Prompt Found, Extract:
+          ```yaml
+          skill:
+            name: ""
+            source_url: ""
+            source_type: "claude_skill | agent_framework | prompt_library | workflow_template"
+            description: ""
+            capabilities: []
+            domain_fit:
+              relevance: "direct | adaptable | inspirational"
+              adaptation_effort: "none | low | medium | high"
+            technical:
+              format: "markdown | yaml | python | json"
+              framework: "claude_code | langchain | crewai | autogen | standalone"
+              dependencies: []
+              self_contained: true/false
+            quality:
+              has_documentation: true/false
+              has_examples: true/false
+              community_usage: "N stars/downloads/forks"
+            reusability:
+              can_fork: true/false
+              can_import: true/false
+              prompt_extractable: true/false
+              workflow_extractable: true/false
+            fills_gaps: []
+          ```
+
+          ## Output Format
+          Return a YAML list of all skills found, sorted by:
+          1. Direct domain relevance
+          2. Adaptation effort (lower is better)
+          3. Quality signals
+          4. Self-contained preference
 
-    # Agent execution logic
+          Include at least {search_depth_count} results.
+        subagent_type: Explore
+        timeout: "{search_timeout}"
+        output_key: skill_results
     execution:
       launch_all_parallel: true
       wait_for_all: true
@@ -613,29 +949,25 @@ phases:
         - cli_results
         - library_results
         - github_results
-
+        - skill_results
     checkpoint:
       id: CP_DEEP_SEARCH
-      name: "Deep Search Complete"
+      name: Deep Search Complete
       blocking: true
       criteria:
-        - at_least_3_agents_succeeded: true
+        - at_least_4_agents_succeeded: true
         - total_tools_found: ">= 10"
       on_partial_failure:
-        action: "Continue with available results, note failures"
-
-  # ─────────────────────────────────────────────────────────────────────────────
-  # PHASE 2: COMPARATIVE EVALUATION (Tiers Relativos)
-  # ─────────────────────────────────────────────────────────────────────────────
+        action: Continue with available results, note failures
   - id: phase_2
-    name: "COMPARATIVE EVALUATION"
-    purpose: "Evaluate tools using RELATIVE TIERS - comparing tools against each other, not absolute values"
-    duration: "10-15 minutes"
-    mode: "sequential"
-    depends_on: ["phase_1"]
-
-    framework_reference: "data/tool-evaluation-framework.md"
-
+    name: COMPARATIVE EVALUATION
+    task_ref: discover-tools-execute
+    purpose: Evaluate tools using RELATIVE TIERS - comparing tools against each other, not absolute values
+    duration: 10-15 minutes
+    mode: sequential
+    depends_on:
+      - phase_1
+    framework_reference: data/tool-evaluation-framework.md
     philosophy: |
       NÃO EXISTEM VALORES ABSOLUTOS UNIVERSAIS.
 
@@ -644,39 +976,44 @@ phases:
       tiverem 50,000.
 
       SEMPRE comparar ferramentas ENTRE SI, dentro do contexto da pesquisa.
-
     principles:
-      - "COMPARAÇÃO RELATIVA - Tiers baseados nos projetos encontrados, não em números mágicos"
-      - "CONTEXTO DE DOMÍNIO - Ferramentas de nicho competem com ferramentas de nicho"
-      - "FLAGS, NÃO VETOS - Problemas são destacados para decisão humana, não eliminados automaticamente"
-      - "DADOS > HEURÍSTICAS - Normalizar dentro do dataset encontrado"
-
+      - COMPARAÇÃO RELATIVA - Tiers baseados nos projetos encontrados, não em números mágicos
+      - CONTEXTO DE DOMÍNIO - Ferramentas de nicho competem com ferramentas de nicho
+      - FLAGS, NÃO VETOS - Problemas são destacados para decisão humana, não eliminados automaticamente
+      - DADOS > HEURÍSTICAS - Normalizar dentro do dataset encontrado
     steps:
       - id: step_2_1
-        name: "Consolidate & Deduplicate"
-        action: "Merge all subagent results and remove duplicates"
+        name: Consolidate & Deduplicate
+        action: Merge all subagent results (including skills) and local scan, remove duplicates
         logic: |
           all_tools = []
-          for category in [mcp, api, cli, library, github]:
+          for category in [mcp, api, cli, library, github, skill]:
             for tool in category_results:
               tool.category = category
               tool.source_agent = agent_name
               all_tools.append(tool)
 
-          deduplicated = deduplicate_by_name_and_url(all_tools)
+          # Add local skills from Phase 0 scan
+          for local_skill in local_skills_coverage.covers + local_skills_coverage.adaptable:
+            local_skill.category = "local_skill"
+            local_skill.source_agent = "local_scan"
+            all_tools.append(local_skill)
 
+          deduplicated = deduplicate_by_name_and_url(all_tools)
         output:
           consolidated_tools:
-            total: N
-            by_category: { mcp: N, api: N, cli: N, library: N, github: N }
-
-      # ─────────────────────────────────────────────────────────────────────────
-      # STEP 2.2: COLLECT RAW METRICS
-      # ─────────────────────────────────────────────────────────────────────────
+            total: "N"
+            by_category:
+              mcp: "N"
+              api: "N"
+              cli: "N"
+              library: "N"
+              github: "N"
+              skill: "N"
+              local_skill: "N"
       - id: step_2_2
-        name: "Collect Raw Metrics"
-        action: "Gather all metrics for each tool to enable relative comparison"
-
+        name: Collect Raw Metrics
+        action: Gather all metrics for each tool to enable relative comparison
         metrics_to_collect:
           social_proof:
             - github_stars
@@ -689,7 +1026,6 @@ phases:
             - last_commit_days_ago
             - releases_last_year
             - stackoverflow_questions
-
           security:
             - critical_cves_count
             - high_cves_count
@@ -698,150 +1034,141 @@ phases:
             - has_security_policy
             - has_security_audit
             - signed_releases
-
           maturity:
             - age_months
             - major_version
             - has_documentation
             - has_examples
             - has_tests
-
         output:
           raw_metrics_matrix:
-            tool_name: { metric_name: value, ... }
-
-      # ─────────────────────────────────────────────────────────────────────────
-      # STEP 2.3: CALCULATE RELATIVE TIERS (Percentis)
-      # ─────────────────────────────────────────────────────────────────────────
+            tool_name:
+              metric_name: value
+              ...: null
       - id: step_2_3
-        name: "Calculate Relative Tiers"
-        action: "For each dimension, rank tools by percentile within the dataset"
-
+        name: Calculate Relative Tiers
+        action: For each dimension, rank tools by percentile within the dataset
         tier_calculation:
-          method: "percentile"
+          method: percentile
           tiers:
-            tier_1: "Top 20% (above P80)"
-            tier_2: "21-50% (P50 to P80)"
-            tier_3: "51-80% (P20 to P50)"
-            tier_4: "Bottom 20% (below P20)"
-
+            tier_1: Top 20% (above P80)
+            tier_2: 21-50% (P50 to P80)
+            tier_3: 51-80% (P20 to P50)
+            tier_4: Bottom 20% (below P20)
           dimensions:
             social_proof:
               components:
-                - name: "popularity"
-                  metrics: [github_stars, npm_downloads, pypi_downloads]
-                  weight: 0.30
-                - name: "community_health"
-                  metrics: [contributors_count, issues_response_ratio]
+                - name: popularity
+                  metrics:
+                    - github_stars
+                    - npm_downloads
+                    - pypi_downloads
+                  weight: 0.3
+                - name: community_health
+                  metrics:
+                    - contributors_count
+                    - issues_response_ratio
                   weight: 0.25
-                - name: "activity"
-                  metrics: [last_commit_days_ago, releases_last_year]
+                - name: activity
+                  metrics:
+                    - last_commit_days_ago
+                    - releases_last_year
                   weight: 0.25
-                  inverse: [last_commit_days_ago]  # Lower is better
-                - name: "adoption"
-                  metrics: [stackoverflow_questions]
-                  weight: 0.20
-
+                  inverse:
+                    - last_commit_days_ago
+                - name: adoption
+                  metrics:
+                    - stackoverflow_questions
+                  weight: 0.2
             security:
-              note: "For CVEs, compare relative counts - lower tier = fewer CVEs"
+              note: For CVEs, compare relative counts - lower tier = fewer CVEs
               components:
-                - name: "vulnerabilities"
-                  metrics: [critical_cves, high_cves, deps_vulnerabilities]
-                  weight: 0.60
-                  inverse: true  # Fewer is better
-                - name: "practices"
-                  metrics: [has_security_policy, has_security_audit, signed_releases]
-                  weight: 0.40
-
+                - name: vulnerabilities
+                  metrics:
+                    - critical_cves
+                    - high_cves
+                    - deps_vulnerabilities
+                  weight: 0.6
+                  inverse: true
+                - name: practices
+                  metrics:
+                    - has_security_policy
+                    - has_security_audit
+                    - signed_releases
+                  weight: 0.4
             maturity:
               components:
-                - name: "longevity"
-                  metrics: [age_months, major_version]
-                  weight: 0.40
-                - name: "documentation"
-                  metrics: [has_documentation, has_examples]
-                  weight: 0.30
-                - name: "quality"
-                  metrics: [has_tests]
-                  weight: 0.30
-
+                - name: longevity
+                  metrics:
+                    - age_months
+                    - major_version
+                  weight: 0.4
+                - name: documentation
+                  metrics:
+                    - has_documentation
+                    - has_examples
+                  weight: 0.3
+                - name: quality
+                  metrics:
+                    - has_tests
+                  weight: 0.3
         output:
           tier_matrix:
             tool_name:
               social_proof_tier: 1-4
               security_tier: 1-4
               maturity_tier: 1-4
-
-      # ─────────────────────────────────────────────────────────────────────────
-      # STEP 2.4: FLAG ATTENTION ITEMS (Not VETOs)
-      # ─────────────────────────────────────────────────────────────────────────
       - id: step_2_4
-        name: "Flag Attention Items"
-        action: "Identify items that require human attention (flags, not automatic vetos)"
-
+        name: Flag Attention Items
+        action: Identify items that require human attention (flags, not automatic vetos)
         note: |
           FLAGS highlight concerns but do NOT automatically eliminate tools.
           A tool with a flag may still be the best option if all alternatives
           have worse issues.
-
         flags:
           security:
-            - flag: "🔴 CRITICAL_CVE"
-              condition: "Has unpatched critical CVE"
-              action: "Highlight for human review"
-              note: "Check if patch available or workaround exists"
-
-            - flag: "🟠 HIGH_CVE"
-              condition: "Has unpatched high CVE"
-              action: "Compare with alternatives"
-
-            - flag: "🟡 SECURITY_INCIDENT"
-              condition: "History of security incident"
-              action: "Research if resolved"
-
+            - flag: 🔴 CRITICAL_CVE
+              condition: Has unpatched critical CVE
+              action: Highlight for human review
+              note: Check if patch available or workaround exists
+            - flag: 🟠 HIGH_CVE
+              condition: Has unpatched high CVE
+              action: Compare with alternatives
+            - flag: 🟡 SECURITY_INCIDENT
+              condition: History of security incident
+              action: Research if resolved
           maturity:
-            - flag: "🔵 VERY_NEW"
-              condition: "Project < 3 months old"
-              action: "May be innovative or unstable"
-
-            - flag: "🟤 SINGLE_MAINTAINER"
-              condition: "Only 1 contributor"
-              action: "Bus factor concern"
-
-            - flag: "⚪ NO_RECENT_RELEASES"
-              condition: "No releases in 12 months"
-              action: "Check if stable or abandoned"
-
+            - flag: 🔵 VERY_NEW
+              condition: Project < 3 months old
+              action: May be innovative or unstable
+            - flag: 🟤 SINGLE_MAINTAINER
+              condition: Only 1 contributor
+              action: Bus factor concern
+            - flag: ⚪ NO_RECENT_RELEASES
+              condition: No releases in 12 months
+              action: Check if stable or abandoned
           licensing:
-            - flag: "⚫ NO_LICENSE"
-              condition: "No OSS license"
-              action: "Cannot use without license clarification"
-
+            - flag: ⚫ NO_LICENSE
+              condition: No OSS license
+              action: Cannot use without license clarification
         output:
           flagged_tools:
-            tool_name: [list of flags]
-
-      # ─────────────────────────────────────────────────────────────────────────
-      # STEP 2.5: COST-BENEFIT COMPARISON (Relative)
-      # ─────────────────────────────────────────────────────────────────────────
+            tool_name:
+              - list of flags
       - id: step_2_5
-        name: "Cost-Benefit Comparison"
-        action: "Compare paid vs OSS alternatives within same tier"
-
+        name: Cost-Benefit Comparison
+        action: Compare paid vs OSS alternatives within same tier
         principle: |
           NO fixed ROI thresholds.
           Compare tiers between paid and OSS options for same capability.
           Present analysis for human decision.
-
         analysis:
           step_1:
-            action: "Group tools by capability gap they fill"
-
+            action: Group tools by capability gap they fill
           step_2:
-            action: "For each gap, identify paid and OSS options"
-
+            action: For each gap, identify paid and OSS options
           step_3:
-            action: "Compare tiers within each group"
+            action: Compare tiers within each group
             logic: |
               If OSS is Tier 1-2 and Paid is also Tier 1-2:
                 → Prefer OSS (cost advantage)
@@ -849,57 +1176,51 @@ phases:
                 → Present both with tier comparison
               If only Paid options exist:
                 → Present with cost analysis
-
           step_4:
-            action: "Generate comparison table for human decision"
-
+            action: Generate comparison table for human decision
         output:
           cost_comparison:
             by_capability:
               capability_name:
-                oss_options: [{ tool, tier, cost: "free" }]
-                paid_options: [{ tool, tier, cost: "$X/mo" }]
-                recommendation: "OSS Tier 1 exists" | "Consider paid" | "Only paid available"
-
-      # ─────────────────────────────────────────────────────────────────────────
-      # STEP 2.6: RICE SCORING (Relative)
-      # ─────────────────────────────────────────────────────────────────────────
+                oss_options:
+                  - tool: null
+                    tier: null
+                    cost: free
+                paid_options:
+                  - tool: null
+                    tier: null
+                    cost: $X/mo
+                recommendation: OSS Tier 1 exists | Consider paid | Only paid available
       - id: step_2_6
-        name: "RICE Scoring (Relative)"
-        action: "Calculate RICE for each tool, then rank by percentile"
-
+        name: RICE Scoring (Relative)
+        action: Calculate RICE for each tool, then rank by percentile
         rice_components:
           reach:
-            description: "Proportion of squad use cases affected"
-            calculation: "(use_cases_affected / total_use_cases) * 10"
-
+            description: Proportion of squad use cases affected
+            calculation: (use_cases_affected / total_use_cases) * 10
           impact:
-            description: "Magnitude of benefit when implemented"
+            description: Magnitude of benefit when implemented
             assessment:
-              - "Manual work reduction %"
-              - "Quality improvement %"
-              - "Necessity (can squad work without it?)"
-              - "Differentiation value"
-            calculation: "weighted average of assessments"
-
+              - Manual work reduction %
+              - Quality improvement %
+              - Necessity (can squad work without it?)
+              - Differentiation value
+            calculation: weighted average of assessments
           confidence:
-            description: "Certainty about Reach and Impact estimates"
+            description: Certainty about Reach and Impact estimates
             based_on:
-              - "Have we tested similar tool?"
-              - "Quality of documentation"
-              - "Number of reviews/case studies"
-              - "Community activity"
-
+              - Have we tested similar tool?
+              - Quality of documentation
+              - Number of reviews/case studies
+              - Community activity
           effort:
-            description: "Time and resources to implement"
+            description: Time and resources to implement
             components:
               - installation_complexity
               - integration_effort
               - learning_curve
               - maintenance_burden
-
-        formula: "RICE = (Reach * Impact * Confidence) / Effort"
-
+        formula: RICE = (Reach * Impact * Confidence) / Effort
         tier_assignment: |
           Calculate RICE for ALL tools, then assign tiers by percentile:
           - Tier 1: Top 20% RICE scores
@@ -909,60 +1230,44 @@ phases:
 
           NOTE: NO absolute threshold like "RICE > 50 is good".
           A tool with RICE 15 can be Tier 1 if it's the best in the dataset.
-
-      # ─────────────────────────────────────────────────────────────────────────
-      # STEP 2.7: WSJF SCORING (Relative)
-      # ─────────────────────────────────────────────────────────────────────────
       - id: step_2_7
-        name: "WSJF Scoring (Relative)"
-        action: "Calculate WSJF for prioritization, rank by percentile"
-
+        name: WSJF Scoring (Relative)
+        action: Calculate WSJF for prioritization, rank by percentile
         cost_of_delay:
           user_business_value:
-            weight: 0.40
-            question: "How much business value does user gain?"
-            comparison: "Rank tools against each other"
-
+            weight: 0.4
+            question: How much business value does user gain?
+            comparison: Rank tools against each other
           time_criticality:
-            weight: 0.30
-            question: "How urgent is this capability?"
-            context: "Depends on squad project timeline"
-
+            weight: 0.3
+            question: How urgent is this capability?
+            context: Depends on squad project timeline
           risk_reduction:
-            weight: 0.30
-            question: "How much risk does this tool mitigate?"
-
+            weight: 0.3
+            question: How much risk does this tool mitigate?
         job_duration:
-          description: "Estimated implementation time"
+          description: Estimated implementation time
           mapping:
-            1: "< 1 hour"
-            2: "1-4 hours"
-            3: "1 day"
-            5: "1 week"
-            7: "2-4 weeks"
-            10: "> 3 months"
-
-        formula: "WSJF = Cost_of_Delay / Job_Duration"
-
+            "1": < 1 hour
+            "2": 1-4 hours
+            "3": 1 day
+            "5": 1 week
+            "7": 2-4 weeks
+            "10": "> 3 months"
+        formula: WSJF = Cost_of_Delay / Job_Duration
         tier_assignment: |
           Same as RICE - assign tiers by percentile within dataset.
           High WSJF = High value, low effort (prioritize first)
-
-      # ─────────────────────────────────────────────────────────────────────────
-      # STEP 2.8: COMPOSITE TIER & RANKING
-      # ─────────────────────────────────────────────────────────────────────────
       - id: step_2_8
-        name: "Calculate Composite Tier"
-        action: "Combine all dimension tiers into final ranking"
-
+        name: Calculate Composite Tier
+        action: Combine all dimension tiers into final ranking
         composite_calculation:
           weights:
-            rice_tier: 0.30
+            rice_tier: 0.3
             wsjf_tier: 0.25
-            social_proof_tier: 0.20
+            social_proof_tier: 0.2
             security_tier: 0.15
-            maturity_tier: 0.10
-
+            maturity_tier: 0.1
           formula: |
             # Convert tiers to scores (Tier 1 = 4, Tier 2 = 3, etc.)
             tier_to_score = {1: 4, 2: 3, 3: 2, 4: 1}
@@ -974,31 +1279,26 @@ phases:
             elif composite_score >= 2.5: Tier 2
             elif composite_score >= 1.5: Tier 3
             else: Tier 4
-
         decision_matrix:
-          note: "Based on VALUE tier (RICE+WSJF) vs EFFORT tier"
+          note: Based on VALUE tier (RICE+WSJF) vs EFFORT tier
           quadrants:
             quick_wins:
-              criteria: "Tier 1-2 Value AND Tier 1-2 Effort"
-              action: "Implement first"
-
+              criteria: Tier 1-2 Value AND Tier 1-2 Effort
+              action: Implement first
             strategic:
-              criteria: "Tier 1-2 Value AND Tier 3-4 Effort"
-              action: "Plan and schedule"
-
+              criteria: Tier 1-2 Value AND Tier 3-4 Effort
+              action: Plan and schedule
             fill_ins:
-              criteria: "Tier 3-4 Value AND Tier 1-2 Effort"
-              action: "If time permits"
-
+              criteria: Tier 3-4 Value AND Tier 1-2 Effort
+              action: If time permits
             backlog:
-              criteria: "Tier 3-4 Value AND Tier 3-4 Effort"
-              action: "Low priority"
-
+              criteria: Tier 3-4 Value AND Tier 3-4 Effort
+              action: Low priority
         output:
           ranked_tools:
-            - tool: "Tool A"
+            - tool: Tool A
               composite_tier: 1
-              quadrant: "Quick Win"
+              quadrant: Quick Win
               flags: []
               tiers:
                 rice: 1
@@ -1006,41 +1306,39 @@ phases:
                 social_proof: 2
                 security: 1
                 maturity: 2
-
       - id: step_2_9
-        name: "Generate Comparative Report"
-        action: "Create report showing relative rankings and comparisons"
-
+        name: Generate Comparative Report
+        action: Create report showing relative rankings and comparisons
         report_sections:
           tier_distribution:
-            description: "How many tools in each tier"
+            description: How many tools in each tier
             format: |
               Tier 1 (Top 20%): N tools
               Tier 2 (21-50%): N tools
               Tier 3 (51-80%): N tools
               Tier 4 (Bottom 20%): N tools
-
           top_by_category:
-            description: "Best tools per category"
-            categories: [mcp, api, cli, library, github]
-            show: "Top 3 per category with tier info"
-
+            description: Best tools per category
+            categories:
+              - mcp
+              - api
+              - cli
+              - library
+              - github
+            show: Top 3 per category with tier info
           gap_coverage:
-            description: "Which gaps have Tier 1-2 options"
+            description: Which gaps have Tier 1-2 options
             format: |
               ✅ Gap X: Tool A (Tier 1), Tool B (Tier 2)
               ⚠️ Gap Y: Only Tier 3-4 options available
               ❌ Gap Z: No tools found
-
           flags_summary:
-            description: "All flagged items requiring attention"
+            description: All flagged items requiring attention
             format: |
               🔴 CRITICAL_CVE: Tool X, Tool Y
               🔵 VERY_NEW: Tool Z
-
           paid_vs_oss:
-            description: "Comparison for gaps where both exist"
-
+            description: Comparison for gaps where both exist
         output:
           comparative_report:
             tier_distribution: {}
@@ -1049,183 +1347,178 @@ phases:
             gap_coverage: {}
             flags_summary: {}
             paid_vs_oss_analysis: {}
-
     checkpoint:
       id: CP_SCORING
-      name: "Scoring Complete"
+      name: Scoring Complete
       blocking: true
       criteria:
         - all_tools_scored: true
         - rankings_generated: true
-
-  # ─────────────────────────────────────────────────────────────────────────────
-  # PHASE 3: DECISION MATRIX & INTEGRATION PLAN
-  # ─────────────────────────────────────────────────────────────────────────────
   - id: phase_3
-    name: "DECISION MATRIX & INTEGRATION PLAN"
-    purpose: "Create actionable decision matrix and implementation plan"
-    duration: "5-10 minutes"
-    mode: "sequential"
-    depends_on: ["phase_2"]
-
+    name: DECISION MATRIX & INTEGRATION PLAN
+    task_ref: discover-tools-execute
+    purpose: Create actionable decision matrix and implementation plan
+    duration: 5-10 minutes
+    mode: sequential
+    depends_on:
+      - phase_2
     steps:
       - id: step_3_1
-        name: "Generate Decision Matrix"
-        action: "Create impact vs effort quadrant matrix"
+        name: Generate Decision Matrix
+        action: Create impact vs effort quadrant matrix
         matrix:
           quadrants:
             quick_wins:
-              criteria: "impact_score >= 7 AND effort_score >= 7"
-              action: "Implement immediately"
+              criteria: impact_score >= 7 AND effort_score >= 7
+              action: Implement immediately
               priority: 1
-
             strategic:
-              criteria: "impact_score >= 7 AND effort_score < 7"
-              action: "Plan for implementation"
+              criteria: impact_score >= 7 AND effort_score < 7
+              action: Plan for implementation
               priority: 2
-
             fill_ins:
-              criteria: "impact_score < 7 AND effort_score >= 7"
-              action: "Nice to have, low priority"
+              criteria: impact_score < 7 AND effort_score >= 7
+              action: Nice to have, low priority
               priority: 3
-
             avoid:
-              criteria: "impact_score < 7 AND effort_score < 7"
-              action: "Don't implement"
+              criteria: impact_score < 7 AND effort_score < 7
+              action: Don't implement
               priority: 4
-
         output:
           decision_matrix:
             quick_wins: []
             strategic: []
             fill_ins: []
             avoid: []
-
       - id: step_3_2
-        name: "Generate Integration Plan"
-        action: "Create phased implementation plan"
+        name: Generate Integration Plan
+        action: Create phased implementation plan
         plan_structure:
           immediate:
-            timeframe: "Today"
-            criteria: "Quick wins with < 30 min setup"
+            timeframe: Today
+            criteria: Quick wins with < 30 min setup
             actions:
               - tool_name: ""
-                action: "Install via {command}"
-                effort: "5-30 min"
+                action: Install via {command}
+                effort: 5-30 min
                 fills_gap: ""
-
           short_term:
-            timeframe: "This Week"
-            criteria: "Quick wins and high-value strategic"
+            timeframe: This Week
+            criteria: Quick wins and high-value strategic
             actions:
               - tool_name: ""
-                action: "Configure and integrate"
-                effort: "1-4 hours"
+                action: Configure and integrate
+                effort: 1-4 hours
                 fills_gap: ""
-
           medium_term:
-            timeframe: "This Month"
-            criteria: "Strategic tools requiring planning"
+            timeframe: This Month
+            criteria: Strategic tools requiring planning
             actions:
               - tool_name: ""
-                action: "Full integration with squad"
-                effort: "1-2 days"
+                action: Full integration with squad
+                effort: 1-2 days
                 fills_gap: ""
-
           backlog:
-            timeframe: "Future"
-            criteria: "Fill-ins and tools needing evaluation"
+            timeframe: Future
+            criteria: Fill-ins and tools needing evaluation
             actions:
               - tool_name: ""
-                action: "Evaluate when time permits"
+                action: Evaluate when time permits
                 reason: ""
-
         output:
           integration_plan:
             immediate: []
             short_term: []
             medium_term: []
             backlog: []
-
       - id: step_3_3
-        name: "Update Tool Registry"
-        action: "Add discovered tools to global registry"
+        name: Update Tool Registry
+        action: Add discovered tools to global registry
+        policy: internal-first
         updates:
-          - add_to: "mcp_servers"
-            status: "discovered"
+          - add_to: mcp_servers
+            status: discovered
             tools: "{new_mcps}"
-
-          - add_to: "domain_recommendations.{domain}"
+          - add_to: domain_recommendations.{domain}
             essential: "{quick_wins}"
             recommended: "{strategic}"
             optional: "{fill_ins}"
-
-          - add_to: "capability_mapping"
+            constraints:
+              - if capability is internal-covered, keep internal tool as primary
+              - external tools for internal-covered capabilities must be fallback/enhancement
+          - add_to: domain_recommendations.{domain}.skills
+            reusable_local: "{local_skills_that_cover}"
+            adaptable_local: "{local_skills_adaptable}"
+            community: "{community_skills_recommended}"
+          - add_to: capability_mapping
             new_mappings: "{gap_to_tool_mappings}"
-
+          - add_to: skill_recommendations.{domain}
+            local_skills: "{matched_local_skills}"
+            community_skills: "{matched_community_skills}"
+            fork_candidates: "{skills_to_fork}"
         output:
           registry_updates:
-            mcps_added: N
-            apis_added: N
-            mappings_added: N
-
+            mcps_added: "N"
+            apis_added: "N"
+            skills_matched: "N"
+            skills_to_fork: "N"
+            mappings_added: "N"
+            registry_path: .aiox/squad-runtime/discovery/{domain}/tool-registry.yaml
       - id: step_3_4
-        name: "Generate Reports"
-        action: "Create comprehensive documentation"
+        name: Generate Reports
+        action: Create comprehensive documentation
         reports:
-          - name: "Tool Discovery Report"
-            location: "squads/{pack_name}/docs/tool-discovery-report.md"
+          - name: Tool Discovery Report
+            location: .aiox/squad-runtime/discovery/{domain}/tool-discovery-report.md
             sections:
               - executive_summary
+              - internal_coverage_summary
+              - external_search_scope
               - capability_gaps_analyzed
               - tools_discovered_by_category
               - decision_matrix_visual
               - top_recommendations
               - detailed_tool_profiles
-
-          - name: "Integration Plan"
-            location: "squads/{pack_name}/docs/tool-integration-plan.md"
+          - name: Integration Plan
+            location: .aiox/squad-runtime/discovery/{domain}/tool-integration-plan.md
             sections:
               - timeline_overview
               - immediate_actions
               - short_term_plan
               - medium_term_plan
               - backlog
-
-          - name: "Capability-Tools Mapping"
-            location: "squads/{pack_name}/data/capability-tools.yaml"
-            format: "yaml"
-
+          - name: Capability-Tools Mapping
+            location: .aiox/squad-runtime/discovery/{domain}/capability-tools.yaml
+            format: yaml
     checkpoint:
       id: CP_DECISION
-      name: "Decision Matrix Complete"
+      name: Decision Matrix Complete
       blocking: true
       criteria:
         - matrix_generated: true
         - plan_created: true
         - registry_updated: true
         - reports_generated: true
-
-  # ─────────────────────────────────────────────────────────────────────────────
-  # PHASE 4: HANDOFF
-  # ─────────────────────────────────────────────────────────────────────────────
   - id: phase_4
-    name: "HANDOFF"
-    purpose: "Present findings and get approval"
-    duration: "2-5 minutes"
-    mode: "interactive"
-    depends_on: ["phase_3"]
-
+    name: HANDOFF
+    task_ref: discover-tools-execute
+    purpose: Present findings and get approval
+    duration: 2-5 minutes
+    mode: interactive
+    depends_on:
+      - phase_3
     steps:
       - id: step_4_1
-        name: "Present Summary"
-        action: "Display comprehensive discovery summary"
+        name: Present Summary
+        action: Display comprehensive discovery summary
         template: |
           ═══════════════════════════════════════════════════════════════════
           🔧 DEEP TOOL DISCOVERY COMPLETE
           ═══════════════════════════════════════════════════════════════════
 
           DOMAIN: {domain}
+          INTERNAL CAPABILITIES COVERED: {internal_covered_count}
+          EXTERNAL-ONLY GAPS: {external_only_gaps_count}
           CAPABILITY GAPS ANALYZED: {gaps_count}
           TOTAL TOOLS DISCOVERED: {tools_count}
 
@@ -1236,6 +1529,8 @@ phases:
           │ APIs:           {api_count:>3} found │ Top: {top_api}           │
           │ CLIs:           {cli_count:>3} found │ Top: {top_cli}           │
           │ Libraries:      {lib_count:>3} found │ Top: {top_lib}           │
+          │ Skills (Comm):  {skill_count:>3} found │ Top: {top_skill}        │
+          │ Skills (Local): {local_skill_count:>3} match │ Top: {top_local_skill} │
           │ GitHub Projects:{gh_count:>3} found │ Top: {top_gh}            │
           └─────────────────────────────────────────────────────────────────┘
 
@@ -1252,188 +1547,165 @@ phases:
           {quick_wins_table}
 
           REPORTS GENERATED:
-          • squads/{pack_name}/docs/tool-discovery-report.md
-          • squads/{pack_name}/docs/tool-integration-plan.md
-          • squads/{pack_name}/data/capability-tools.yaml
+          • .aiox/squad-runtime/discovery/{domain}/tool-discovery-report.md
+          • .aiox/squad-runtime/discovery/{domain}/tool-integration-plan.md
+          • .aiox/squad-runtime/discovery/{domain}/capability-tools.yaml
 
           ═══════════════════════════════════════════════════════════════════
-
       - id: step_4_2
-        name: "Request Approval"
-        action: "Ask user to approve integration plan"
-        question: "How would you like to proceed?"
+        name: Request Approval
+        action: Ask user to approve integration plan
+        question: How would you like to proceed?
         options:
-          - id: "approve_all"
-            label: "Approve all quick wins - integrate now"
-            action: "Execute immediate actions from integration plan"
-
-          - id: "review_first"
-            label: "Review detailed report first"
-            action: "Open tool-discovery-report.md"
-
-          - id: "selective"
-            label: "Select specific tools to integrate"
-            action: "Present tool list for selection"
-
-          - id: "skip"
-            label: "Skip tool integration for now"
-            action: "Save reports, continue to next phase"
-
-# ═══════════════════════════════════════════════════════════════════════════════
-# OUTPUTS
-# ═══════════════════════════════════════════════════════════════════════════════
-
+          - id: approve_all
+            label: Approve all quick wins - integrate now
+            action: Execute immediate actions from integration plan
+          - id: review_first
+            label: Review detailed report first
+            action: Open .aiox/squad-runtime/discovery/{domain}/tool-discovery-report.md
+          - id: selective
+            label: Select specific tools to integrate
+            action: Present tool list for selection
+          - id: skip
+            label: Skip tool integration for now
+            action: Save reports, continue to next phase
 outputs:
   primary:
-    - name: "Tool Discovery Report"
-      location: "squads/{pack_name}/docs/tool-discovery-report.md"
-      description: "Comprehensive research findings with all discovered tools"
-
-    - name: "Integration Plan"
-      location: "squads/{pack_name}/docs/tool-integration-plan.md"
-      description: "Prioritized implementation timeline"
-
-    - name: "Decision Matrix"
-      location: "Embedded in tool-discovery-report.md"
-      description: "Impact vs Effort quadrant visualization"
-
+    - name: Tool Discovery Report
+      location: .aiox/squad-runtime/discovery/{domain}/tool-discovery-report.md
+      description: Comprehensive research findings with all discovered tools
+    - name: Integration Plan
+      location: .aiox/squad-runtime/discovery/{domain}/tool-integration-plan.md
+      description: Prioritized implementation timeline
+    - name: Decision Matrix
+      location: Embedded in .aiox/squad-runtime/discovery/{domain}/tool-discovery-report.md
+      description: Impact vs Effort quadrant visualization
   secondary:
-    - name: "Capability-Tools Mapping"
-      location: "squads/{pack_name}/data/capability-tools.yaml"
-      description: "Squad-specific capability to tool mapping"
-
-    - name: "Updated Tool Registry"
-      location: "data/tool-registry.yaml"
-      description: "Global registry updated with discoveries"
-
-# ═══════════════════════════════════════════════════════════════════════════════
-# HEURISTICS
-# ═══════════════════════════════════════════════════════════════════════════════
-
+    - name: Capability-Tools Mapping
+      location: .aiox/squad-runtime/discovery/{domain}/capability-tools.yaml
+      description: Domain capability to tool mapping
+    - name: Updated Tool Registry
+      location: .aiox/squad-runtime/discovery/{domain}/tool-registry.yaml
+      description: Runtime registry snapshot updated with discoveries
 heuristics:
+  - id: SC_DCV_001
+    name: Domain Context Validation Complete
+    phase: phase_minus_2
+    blocking: true
+    criteria:
+      - canonical_scope_extracted: true
+      - scope_from_name_only: false
+      - ambiguity_gate_passed: true
+    notes:
+      - Discovery is forbidden when scope is inferred from name/slug only
+      - If canonical sources conflict, pause and request clarification
+  - id: SC_INT_001
+    name: Internal-First Coverage Complete
+    phase: phase_minus_1
+    blocking: true
+    criteria:
+      - internal_library_loaded: true
+      - capabilities_mapped_to_internal: true
+      - external_scope_restricted_to_internal_gaps: true
+    notes:
+      - External search is forbidden before internal coverage mapping
+      - Internal tools remain primary when coverage exists
   - id: SC_TLD_001
-    name: "Tool Discovery Complete"
-    phase: "phase_3"
+    name: Tool Discovery Complete
+    phase: phase_3
     blocking: true
     criteria:
-      - parallel_agents_executed: ">= 3 of 5"
-      - tools_discovered: ">= 5"  # At least some tools found
+      - parallel_agents_executed: ">= 4 of 6"
+      - tools_discovered: ">= 5"
       - tier_distribution_calculated: true
       - comparative_report_generated: true
     notes:
-      - "No minimum tool count threshold - even 5 tools can be valuable if ranked"
-      - "Quality is determined by relative ranking, not absolute counts"
-
+      - No minimum tool count threshold - even 5 tools can be valuable if ranked
+      - Quality is determined by relative ranking, not absolute counts
   - id: SC_TLD_002
-    name: "Relative Ranking Complete"
-    phase: "phase_2.step_2_3"
+    name: Relative Ranking Complete
+    phase: phase_2.step_2_3
     blocking: true
     criteria:
       - all_tools_have_tiers: true
       - percentile_calculation_done: true
     notes:
-      - "NO absolute thresholds - all tools get tiers based on comparison"
-      - "Even Tier 4 tools are included - they may be the only option"
-
+      - NO absolute thresholds - all tools get tiers based on comparison
+      - Even Tier 4 tools are included - they may be the only option
   - id: SC_TLD_003
-    name: "Flags Identified"
-    phase: "phase_2.step_2_4"
+    name: Flags Identified
+    phase: phase_2.step_2_4
     blocking: false
     criteria:
       - security_flags_checked: true
       - maturity_flags_checked: true
       - license_flags_checked: true
     notes:
-      - "Flags are for ATTENTION, not automatic elimination"
-      - "A tool with flags may still be recommended if best in tier"
-
+      - Flags are for ATTENTION, not automatic elimination
+      - A tool with flags may still be recommended if best in tier
   - id: SC_TLD_004
-    name: "Cost Comparison Done"
-    phase: "phase_2.step_2_5"
+    name: Cost Comparison Done
+    phase: phase_2.step_2_5
     blocking: false
     criteria:
       - paid_vs_oss_grouped: true
       - tier_comparison_per_gap: true
     notes:
-      - "NO fixed ROI thresholds"
-      - "Present tier comparison for human decision"
-      - "OSS preferred when same tier as paid"
-
+      - NO fixed ROI thresholds
+      - Present tier comparison for human decision
+      - OSS preferred when same tier as paid
+  - id: SC_TLD_005a
+    name: Skills Coverage Complete
+    phase: phase_0.step_0_2b + phase_1.agent_skill
+    blocking: false
+    criteria:
+      - local_skills_scanned: true
+      - local_squads_scanned: true
+      - community_skills_searched: true
+      - skills_matched_to_gaps: true
+    notes:
+      - Local skills are ALWAYS checked first (zero cost, immediate value)
+      - Community skills supplement what local doesn't cover
+      - Skills with 'adaptable' status are fork candidates
+      - A skill that covers a gap can eliminate external tool need for that gap
   - id: SC_TLD_005
-    name: "Decision Matrix Generated"
-    phase: "phase_2.step_2_8"
+    name: Decision Matrix Generated
+    phase: phase_2.step_2_8
     blocking: true
     criteria:
-      - quadrants_assigned: true  # Quick Win, Strategic, Fill-in, Backlog
+      - quadrants_assigned: true
       - recommendations_by_gap: true
     notes:
-      - "Quadrants based on relative tiers, not absolute scores"
-      - "All tools get a quadrant - none automatically excluded"
-
-# ═══════════════════════════════════════════════════════════════════════════════
-# SEARCH DEPTH CONFIGURATION
-# ═══════════════════════════════════════════════════════════════════════════════
-
+      - Quadrants based on relative tiers, not absolute scores
+      - All tools get a quadrant - none automatically excluded
 search_depth_config:
   quick:
     results_per_category: 5
     timeout_minutes: 5
-    use_case: "Fast exploration"
-
+    use_case: Fast exploration
   standard:
     results_per_category: 15
     timeout_minutes: 10
-    use_case: "Normal squad creation"
-
+    use_case: Normal squad creation
   exhaustive:
     results_per_category: 30
     timeout_minutes: 20
-    use_case: "Enterprise squads, complex domains"
-
-# ═══════════════════════════════════════════════════════════════════════════════
-# ERROR HANDLING
-# ═══════════════════════════════════════════════════════════════════════════════
-
+    use_case: Enterprise squads, complex domains
 error_handling:
+  ambiguous_or_unknown_scope:
+    action: Halt discovery and request clarification
+    retry: false
+  canonical_scope_conflict:
+    action: Log contradictions and request human resolution before proceeding
+    retry: false
   agent_timeout:
-    action: "Continue with other agents, mark as partial"
+    action: Continue with other agents, mark as partial
     retry: false
-
   agent_failure:
-    action: "Log error, continue with other agents"
+    action: Log error, continue with other agents
     retry_count: 1
-
   no_tools_found:
-    action: "Expand search to adjacent domains"
-    fallback: "Mark as 'custom development needed'"
-
+    action: Expand search to adjacent domains
+    fallback: Mark as 'custom development needed'
   scoring_failure:
-    action: "Use default scores, flag for manual review"
-
-# ═══════════════════════════════════════════════════════════════════════════════
-# METADATA
-# ═══════════════════════════════════════════════════════════════════════════════
-
-metadata:
-  version: "2.0"
-  created: "2026-02-03"
-  updated: "2026-02-03"
-  author: "squad-chief"
-  lines: "900+"
-  purpose: "Deep parallel tool discovery with RELATIVE TIERS - comparing tools against each other, not absolute values"
-  philosophy: "Não existem valores absolutos universais. Um projeto com 30 stars pode ser excelente se for o melhor do nicho."
-  changelog:
-    v2.0:
-      - "MAJOR: Replaced absolute thresholds with RELATIVE TIERS"
-      - "Tools compared by PERCENTILE within dataset (Top 20%, 21-50%, etc.)"
-      - "Replaced VETOs with FLAGS requiring human attention"
-      - "No tool automatically eliminated - Tier 4 may still be best option"
-      - "Cost comparison: tier-based, no fixed ROI thresholds"
-      - "RICE/WSJF scores ranked relatively, not against fixed values"
-      - "Framework reference: data/tool-evaluation-framework.md v2.0"
-    v1.1:
-      - "Added RICE Framework scoring"
-      - "Added WSJF Framework scoring"
-      - "Added Security/Social Proof gates (replaced in v2.0)"
-    v1.0:
-      - "Initial release with 5 parallel sub-agents"
+    action: Use default scores, flag for manual review