Skip to main content

covariant-modification

Unified skill modification with covariant transport, Darwin Gödel Machine evolution, and MCP Tasks self-rewriting. GF(3) conserved.

소스 정보

저장소
plurigrid/asi
최근 소스 활동
2026년 6월 10일 11:55
감지된 SKILL.md 언어
영어
스타
67
포크
12

설치 방법

기본적으로 소스를 먼저 확인하는 Prompt가 선택됩니다. 직접 명령으로 전환하거나 로컬 사본을 다운로드할 수도 있습니다.

소스 파일 검토

설치 여부를 결정하기 전에 SKILL.md와 SkillsMP에 표시된 보조 파일을 읽어 보세요.

파일 탐색기
2 개 파일

SKILL.md 표시 중

SKILL.md
소스 지침 · 읽기 전용 미리보기
name
covariant-modification
description
Unified skill modification with covariant transport, Darwin Gödel Machine evolution, and MCP Tasks self-rewriting. GF(3) conserved.
trit
0
color
#26D826
# Covariant Modification Skill **Trit**: 0 (ERGODIC - coordinator) **Color**: Green (#26D826) ## Overview **Covariant Modification** unifies three skill patterns for safe, structure-preserving self-modification: | Component Skill | Trit | Role | Pattern | |-----------------|------|------|---------| | `codex-self-rewriting` | +1 | Generator | Lisp-machine self-modification via MCP Tasks | | `self-evolving-agent` | 0 | Coordinator | Darwin Gödel Machine evolution loops | | `covariant-fibrations` | -1 | Validator | Type families respect directed morphisms | **GF(3)**: (+1) + (0) + (-1) = 0 ✓ ## Core Concept: Covariant Transport When skill `A` modifies itself, dependent skills `B` must transform **covariantly**: ``` modify_A Skill_A ─────────────→ Skill_A' │ │ uses │ COVARIANT │ uses' │ TRANSPORT │ ↓ ↓ Skill_B ─────────────→ Skill_B' transport_f ``` ### Agda Definition ```agda -- Skill fibration over dependency base skill-fibration : (Base : SkillGraph) → (Fiber : Base → SkillVersion) → Type -- Covariant transport along modification morphisms cov-transport : {A A' : Skill} {P : SkillDeps A → Type} → (f : Modification A A') → P (deps A) → P (deps A') -- Functoriality cov-comp : ∀ (f : Mod A A') (g : Mod A' A'') → cov-transport (g ∘ f) ≡ cov-transport g ∘ cov-transport f ``` ## MCP Tasks State Machine From `codex-self-rewriting`: ``` ┌─────────────┐ │ working │ LIVE (+1) │ (modify) │ └──────┬──────┘ │ ┌────────────┼────────────┐ ↓ ↓ ↓ ┌─────────────┐ ┌─────────────┐ ┌─────────────┐ │ completed │ │input_required│ │ failed │ │ BACKFILL(-1)│ │ VERIFY (0) │ │ BACKFILL(-1)│ └─────────────┘ └─────────────┘ └─────────────┘ ``` ## Darwin Gödel Machine Integration ```python class CovariantDGM(DarwinGodelMachine): """Darwin Gödel Machine with covariant skill transport.""" def __init__(self, skill_fibration: SkillFibration, ...): super().__init__(...) self.fibration = skill_fibration def mutate(self, agent: Agent) -> Agent: """Mutate agent while preserving fibration structure.""" new_code = self.mutator(agent) # Transport dependent skills covariantly modified_deps = {} for dep_skill in self.fibration.dependencies(agent): modified_deps[dep_skill] = self.fibration.transport( modification=Diff(agent.code, new_code), target=dep_skill ) return Agent( code=new_code, dependencies=modified_deps, generation=self.generation ) ``` ## Multi-Agent Sheaf Gluing ```python class CovariantModificationSheaf: """Sheaf ensuring consistent modifications across agents.""" def glue(self, local_mods: Dict[Agent, Modification]) -> GlobalMod: """Glue compatible local modifications into global section.""" for a1, a2 in combinations(local_mods.keys(), 2): overlap = self.overlap(a1, a2) if overlap: r1 = self.restrict(local_mods[a1], overlap) r2 = self.restrict(local_mods[a2], overlap) if not self.compatible(r1, r2): raise CovarianceViolation(a1, a2, overlap) return self.colimit(local_mods) ``` ## Triadic Modification Operators | Trit | Effect | Operator | Example | |------|--------|----------|---------| | +1 | **Generative** | Create new structure | Add skill capability | | 0 | **Neutral** | Refactor/reorganize | Rename function | | -1 | **Destructive** | Remove/simplify | Delete unused code | **Conservation Law**: ``` Σ trit(modification_i) ≡ 0 (mod 3) ``` ## GF(3) Triads ``` covariant-fibrations (-1) ⊗ covariant-modification (0) ⊗ codex-self-rewriting (+1) = 0 ✓ temporal-coalgebra (-1) ⊗ covariant-modification (0) ⊗ self-evolving-agent (+1) = 0 ✓ sheaf-cohomology (-1) ⊗ covariant-modification (0) ⊗ gay-mcp (+1) = 0 ✓ ``` ## Commands ```bash # Verify covariant modification just covariant-verify skill=my-skill mod=v1.1 # Run DGM evolution with covariance check just dgm-evolve --covariant --generations=50 # Check GF(3) conservation just gf3-audit modified-skills/ # Apply modification with transport just covariant-modify skill=target mod=change.diff ``` ## Related Skills - `covariant-fibrations` (-1): Type transport validation - `self-evolving-agent` (0): DGM evolution patterns - `codex-self-rewriting` (+1): MCP Tasks self-modification - `bisimulation-game` (-1): Observational equivalence verification ## See Also - [Covariant Fibrations in Directed Type Theory](https://arxiv.org/abs/2211.01602) - [Darwin Gödel Machine](https://hf.co/papers/2505.22954) - [MCP Tasks Specification](https://modelcontextprotocol.io/specification/draft/basic/utilities/tasks) ## CT lattice atlas Part of: `para-mensch-commons` (CT lattice family).
GitHub에서 보기