recursive-mode

What problem does it solve?

Coordinating multi-phase AI-assisted software development in a way that is reproducible, auditable, and durable can be error-prone when relying on chat history alone. recursive-mode provides a file-backed, stage-gated workflow that stores requirements, plans, implementation evidence, tests, and memory in the repository, enabling consistent resumption, auditability, and long-term knowledge retention.

Core Features & Use Cases

• File-backed runs: Each run stores requirements, plan, implementation evidence, tests, and memory artifacts inside the repo for reliability and auditability.
• Phase governance: Enforces the pipeline through draft -> audit -> repair -> re-audit -> pass -> lock gates to ensure quality and traceability.
• Subskill and memory integration: Supports subskills, addenda, and a memory plane to improve future runs and enable durable knowledge.
• Memory promotion: Phase 8 records run-local skill usage and promotes durable lessons into memory surfaces for reuse.
• Routing and delegation: Integrates with a canonical router to route bounded delegated work through external CLIs/models when configured.

Quick Start

Bootstraps a reusable .recursive scaffold and starts a run with a short command like Implement the run.

request: > Automates and enforces a file-backed, phase-gated workflow for AI-assisted repository tasks. Operates across requirements, plan, implementation, testing, review, and memory with addenda support and durable memory updates. Ensures deterministic diff-basis, audit loops, and routed delegation with subagent verification.