Mark/mesh-seed

Fork 0

[Phase 5.6] Meta-Factory (Factory of Factories) #68

New issue

Closed

opened 2026-04-01 02:00:34 +00:00 by Mark · 1 comment

Mark commented

2026-04-01 02:00:34 +00:00

Owner

Overview

Enable the mesh to create and evolve its own factories - a "factory of factories" that can generate new factory types.

Problem Statement

Current factories are fixed:

worker_factory - creates worker definitions
skill_factory - creates skills
capability_factory - creates capabilities
prompt_factory - creates prompts
contract_factory - creates contracts

We cannot:

Create new factory types
Evolve factory mutation strategies
Discover new mutation patterns

Proposed Solution

5.6.1: Meta-Factory Architecture

Create capabilities/_muscle/meta_factory/:

# capabilities/_muscle/meta_factory/template.yaml
capability:
  id: meta_factory_v1
  class: muscle
  
skills:
  - create_factory
  - mutate_factory
  - analyze_factory_effectiveness
  
contracts:
  - intent_type: factory_creation
    skill_family: meta_factory
    input_schema:
      - target_domain: str  # What the factory should create
      - mutation_strategies: list[str]
      - evaluation_criteria: dict
    output_schema:
      - factory_id: str
      - template_path: str

5.6.2: Factory Specification

@dataclass
class FactorySpec:
    """Specification for a new factory."""
    
    factory_id: str
    target_domain: str  # e.g., "prompts", "contracts", "workflows"
    mutation_strategies: list[MutationStrategy]
    evaluation_criteria: dict
    template_schema: dict
    variant_schema: dict

5.6.3: Factory Creation Skill

def create_factory(
    inputs: dict,
    context: dict
) -> dict:
    """Create a new factory capability.
    
    Steps:
    1. Analyze target domain (what should factory create?)
    2. Design mutation strategies (how to vary outputs?)
    3. Generate factory template.yaml
    4. Generate factory run.py (using ImplementationGenerator)
    5. Register factory in capabilities/
    6. Return factory_id
    """

5.6.4: Factory Evolution

Meta-Factory monitors factory effectiveness
        ↓
Identify underperforming factories
        ↓
Generate factory variants:
  - New mutation strategies
  - Different evaluation criteria
  - Modified template schemas
        ↓
Test factory variants in parallel
        ↓
Promote better-performing factory

Example: Creating a Workflow Factory

# Request to meta_factory
{
    "intent_type": "factory_creation",
    "payload": {
        "target_domain": "workflows",
        "mutation_strategies": [
            "add_step",
            "remove_step", 
            "reorder_steps",
            "parallelize_steps",
            "add_checkpoint"
        ],
        "evaluation_criteria": {
            "success_rate_weight": 0.5,
            "latency_weight": 0.3,
            "cost_weight": 0.2
        }
    }
}

# Output: workflow_factory created
{
    "factory_id": "workflow_factory_v1",
    "template_path": "capabilities/_muscle/workflow_factory/template.yaml",
    "status": "ready"
}

Deliverables

capabilities/_muscle/meta_factory/ - Meta-factory capability
create_factory skill - Generate new factory
mutate_factory skill - Evolve existing factory
analyze_factory_effectiveness skill - Track factory performance
FactorySpec dataclass - Factory specification
Factory template generation
Factory implementation generation
Integration with LineageTracker

Factory Evolution Metrics

Metric	Description
Variant success rate	% of variants that get promoted
Mean championship score	Average score of factory's variants
Mutation diversity	Variety of mutation strategies used
Domain coverage	Range of problems factory addresses

Success Criteria

Can create new factory for any domain
New factory generates valid variants
Factory variants compete in championships
Factories can be mutated and improved
Meta-factory can create 2+ new factory types

Dependencies

#64 (Factory Learning) - analytics for factory effectiveness
#67 (Lineage Tracking) - track factory genealogy

Estimated Duration

2 weeks

Parent: #62 (Phase 5 Meta)
Uses: capabilities/_shared/base.py (MutationRegistry)

## Overview Enable the mesh to create and evolve its own factories - a "factory of factories" that can generate new factory types. ## Problem Statement Current factories are fixed: - `worker_factory` - creates worker definitions - `skill_factory` - creates skills - `capability_factory` - creates capabilities - `prompt_factory` - creates prompts - `contract_factory` - creates contracts We cannot: - Create new factory types - Evolve factory mutation strategies - Discover new mutation patterns ## Proposed Solution ### 5.6.1: Meta-Factory Architecture Create `capabilities/_muscle/meta_factory/`: ```yaml # capabilities/_muscle/meta_factory/template.yaml capability: id: meta_factory_v1 class: muscle skills: - create_factory - mutate_factory - analyze_factory_effectiveness contracts: - intent_type: factory_creation skill_family: meta_factory input_schema: - target_domain: str # What the factory should create - mutation_strategies: list[str] - evaluation_criteria: dict output_schema: - factory_id: str - template_path: str ``` ### 5.6.2: Factory Specification ```python @dataclass class FactorySpec: """Specification for a new factory.""" factory_id: str target_domain: str # e.g., "prompts", "contracts", "workflows" mutation_strategies: list[MutationStrategy] evaluation_criteria: dict template_schema: dict variant_schema: dict ``` ### 5.6.3: Factory Creation Skill ```python def create_factory( inputs: dict, context: dict ) -> dict: """Create a new factory capability. Steps: 1. Analyze target domain (what should factory create?) 2. Design mutation strategies (how to vary outputs?) 3. Generate factory template.yaml 4. Generate factory run.py (using ImplementationGenerator) 5. Register factory in capabilities/ 6. Return factory_id """ ``` ### 5.6.4: Factory Evolution ``` Meta-Factory monitors factory effectiveness ↓ Identify underperforming factories ↓ Generate factory variants: - New mutation strategies - Different evaluation criteria - Modified template schemas ↓ Test factory variants in parallel ↓ Promote better-performing factory ``` ## Example: Creating a Workflow Factory ```python # Request to meta_factory { "intent_type": "factory_creation", "payload": { "target_domain": "workflows", "mutation_strategies": [ "add_step", "remove_step", "reorder_steps", "parallelize_steps", "add_checkpoint" ], "evaluation_criteria": { "success_rate_weight": 0.5, "latency_weight": 0.3, "cost_weight": 0.2 } } } # Output: workflow_factory created { "factory_id": "workflow_factory_v1", "template_path": "capabilities/_muscle/workflow_factory/template.yaml", "status": "ready" } ``` ## Deliverables - [ ] `capabilities/_muscle/meta_factory/` - Meta-factory capability - [ ] `create_factory` skill - Generate new factory - [ ] `mutate_factory` skill - Evolve existing factory - [ ] `analyze_factory_effectiveness` skill - Track factory performance - [ ] `FactorySpec` dataclass - Factory specification - [ ] Factory template generation - [ ] Factory implementation generation - [ ] Integration with LineageTracker ## Factory Evolution Metrics | Metric | Description | |--------|-------------| | Variant success rate | % of variants that get promoted | | Mean championship score | Average score of factory's variants | | Mutation diversity | Variety of mutation strategies used | | Domain coverage | Range of problems factory addresses | ## Success Criteria - [ ] Can create new factory for any domain - [ ] New factory generates valid variants - [ ] Factory variants compete in championships - [ ] Factories can be mutated and improved - [ ] Meta-factory can create 2+ new factory types ## Dependencies - #64 (Factory Learning) - analytics for factory effectiveness - #67 (Lineage Tracking) - track factory genealogy ## Estimated Duration 2 weeks ## Related - Parent: #62 (Phase 5 Meta) - Uses: `capabilities/_shared/base.py` (MutationRegistry)

Mark added the

high-priority

phase-5

self-building

labels

2026-04-01 03:19:23 +00:00

Mark closed this issue

2026-04-02 16:54:38 +00:00

Mark referenced this issue from a commit

2026-04-02 16:54:39 +00:00

Complete Phase 5 Self-Building Infrastructure

Mark commented

2026-04-02 16:54:44 +00:00

Author

Owner

✅ Implementation Complete

Commit: 3736dfb

Components Delivered:

capabilities/_muscle/meta_factory/ - Meta-Factory capability with 3 skills
evolution/lineage_tracker.py - Genealogy tracking
evolution/trait_extractor.py - Trait extraction from AST
evolution/capability_fusion.py - Capability fusion with 3 strategies
evolution/factory_analytics.py - Factory analytics

Tests: 845 passing (804 unit + 41 integration)

Key Features:

create_factory: Generate new factory types from domain specs
mutate_factory: Evolve existing factories with new strategies
analyze_factory_effectiveness: Query FactoryAnalytics for insights
Human approval gates for new factory creations
Lineage integration tracks factory evolution

API Endpoints: /v1/lineage/* for genealogy queries

✅ **Implementation Complete** **Commit:** 3736dfb **Components Delivered:** - `capabilities/_muscle/meta_factory/` - Meta-Factory capability with 3 skills - `evolution/lineage_tracker.py` - Genealogy tracking - `evolution/trait_extractor.py` - Trait extraction from AST - `evolution/capability_fusion.py` - Capability fusion with 3 strategies - `evolution/factory_analytics.py` - Factory analytics **Tests:** 845 passing (804 unit + 41 integration) **Key Features:** - `create_factory`: Generate new factory types from domain specs - `mutate_factory`: Evolve existing factories with new strategies - `analyze_factory_effectiveness`: Query FactoryAnalytics for insights - Human approval gates for new factory creations - Lineage integration tracks factory evolution **API Endpoints:** `/v1/lineage/*` for genealogy queries

Mark referenced this issue

2026-04-02 18:07:31 +00:00

[Phase 5.4] Automated Deployment Pipeline #66

No milestone

No project

No assignees

1 participant

Notifications

Due date

The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set.

Reference

Mark/mesh-seed#68

No description provided.

Rows
Columns