synthesis-context-lifecycle

What problem does it solve?

Long-running AI collaborations accumulate unwieldy context files that degrade agent performance and create continuity failures across sessions. This Skill prescribes a three-tier context architecture (working memory, semantic reference, episodic archive) and a disciplined archival-first workflow so projects remain discoverable, performant, and auditable.

Core Features & Use Cases

• Three-tier architecture: Separate CONTEXT.md for session working memory, REFERENCE.md for stable facts, and sessions/YYYY-MM.md for monthly archives to limit growth and improve retrieval.
• Archival-first protocol: Two-phase commit pattern that writes archives before removing source content, with verification steps and strict per-repo commit scoping to avoid lost context.
• Operational templates & budgets: Clear templates and line-count budgets (CONTEXT.md ≤150 lines, REFERENCE.md ≤300 lines) plus migration guidance for projects of varying sizes.
• Commit protocol & scope rules: Enforce committing and pushing context changes at point-of-modification and never running workspace-wide sweep commits, making context changes visible across machines immediately.
• Use Case: Maintain continuity for a multi-week engineering project where session logs, decisions, and active tasks must be kept compact, auditable, and shareable between human and AI collaborators.

Quick Start

Open the project's CONTEXT.md, extract session logs older than one week to sessions/YYYY-MM.md, move stable facts into REFERENCE.md, trim CONTEXT.md to the working-memory budget, then commit and push only the modified files.