Перейти к содержанию

Self-improving agentic orchestration runtime

Дата: 2026-08-29 Репозиторий: https://github.com/watt-mind/factory Уверенность: в карантине Модель: claude-haiku-4-5-20251001 Промпт версия: v2.0 Источник: файл: README локация: не указана цитата: "The factory that builds software — and itself. A runtime for self-improving agentic loops: the tracker is the control plane, git is the truth, CI is the gate."

What Changes in the Ecosystem

Factory introduces an orchestration layer that decouples coding agents from process responsibility: agents become stateless workers, while the runtime becomes the standing control loop. This shifts the market model from "agent produces diff → human merges" to "agent produces diff → runtime verifies → human reviews provable artifact". The ecosystem gains a reusable primitive for agent-authored code pipelines across any GitHub/git-based repository, reducing friction for autonomous contribution workflows.

Reasoning

Factory is genuinely novel as an architectural layer: it solves the process gap around coding agents with a purpose-built runtime that enforces verification gates and prevents agent-state fragmentation. Maturity is low (v0.1.0, young, sharp edges) despite active self-use; the badge showing autonomous PR merges indicates real usage momentum but not production hardening yet. State_value is "Growing" because the project is being actively maintained and used by its authors at scale (742 PRs/month), with clear adoption signals.

Maturity x Novelty

Maturity: 2/5 Novelty: 4/5

Self-Check (CoVe)

Cross-validation (README vs manifest/files): The README claims factory is "a runtime for self-improving agentic loops" with a specific architecture: tracker as control plane, git as truth, CI as gate. The manifest confirms this via package.json keywords (agent, agentic, dispatch, orchestration, ci, coding-agent) and files: orchestrator/, event-runtime/, lib/runners/, and tools/ directories support the claimed event-driven routing and dispatch system. The badge badge shows 742 PRs merged autonomously this month, confirming active self-use. The root files include orchestrator scripts (linear-reaper, dispatch, merge, janitor) matching the claimed loop stages (Triage → agent-ready → PR → Done). However, the version is 0.1.0 and README states "Young software, used in earnest every day — expect sharp edges", indicating prototype-to-early-production transition rather than hardened deployment. Подтверждено: Да

Novelty checklist: New protocol: Partially yes — factory establishes a formal coordination protocol between human ticketing systems, AI agents, and CI gates. New standard: Not explicitly; no formal spec is claimed. New architectural layer: Yes — the runtime layer that sits between coding agents (Claude, Gemini, Cursor) and git/CI as a deterministic control plane is novel; it abstracts agent dispatch, work-tree isolation, and verification-before-merge as a composable layer. New way of market interaction: Yes — it enables autonomous agent contributions to software that survive human code review, shifting from "agent as code suggester" to "agent as author in verified workflow". Проходит: Да

Falsifiable Hypothesis

Если права: In 12 months, factory reaches v1.0, receives adoption by 3+ public projects beyond watt-mind, and the badge reports 1000+ PRs merged autonomously across multiple repositories with zero critical security incidents. Если ошиблась: In 12 months, the project remains at v0.x with only watt-mind's internal use, incident reports of agent commits bypassing intended review gates, or core abstractions (ControlPlane adapter, event-runtime) are rewritten due to architectural mismatch discovered in production.

Оценка Claude

Правка человека

Мнение Ольги

История оценок

  • 2026-09-12 - VALIDATED_SHIFT: карантин пройден (14+ дней), репозиторий активен - promote_candidates
  • 2026-08-29 - CANDIDATE: первая оценка

Связи

  • [[Верификация как встроенная архитектура доверия 2026-07-23]]
  • [[MCP как универсальный протокол агентной интеграции 2026-06-14]]
  • [[Оркестрация множества коммерческих агентов 2026-07-23]]
  • [[Human Verification Embedded in Agent Loops 2026-08-04]]
  • [[Открытые протоколы координации агентов 2026-06-25]]