What problem does it solve?
LLM conversation context is ephemeral: summarization, session changes, and token limits can cause mid-task progress, decisions, and state for AI agents generating software artifacts to be lost permanently, forcing work to be restarted from scratch.
Core Features & Use Cases
- Skeleton-First Write: Creates the full artifact structure upfront with progress tracking, so work is recoverable even if interrupted immediately after starting.
- Section-Atomic Saves: Persists each completed section to disk immediately, so no completed work is lost if context is lost mid-generation.
- Resume-on-Entry: Automatically detects in-progress artifacts and picks up exactly where the agent left off, recovering all prior decisions from the artifact itself instead of conversation memory.
- Context Canary Gate: Validates every write operation against the artifact's stored progress to prevent duplicate or out-of-order section generation after summarization.
- Use Case: Any agent in the Myrmion AI Factory SDLC workflow (SETUP, CODESIGN, BLUEPRINT, IMPLEMENT, DEVOPS, QA) generating artifacts like design documents, specifications, test plans, or infrastructure configurations.
Quick Start
Use the factory-incremental-persistence skill to automatically save all agent artifact work incrementally so no progress is lost during context summarization or session interruptions.
