Cross-Session Memory Bridge

The Problem

The hard part of cross-session memory is not storing more, but remembering at the right moment. When a chat ends, project background, confirmed budget, scope, and the last working state can disappear; the next day the user must re-explain context, and the model may answer without relevant history. Loading every conversation with all prior history, however, creates noise and cross-project contamination.

How It Works

context-bridge-pro runs as a passive background skill without requiring user commands. It works through three core steps:
- Project identification: it matches the current conversation to a project using names, keywords, or automatically accumulated aliases, then reads the matching context.md and weaves the information into the response naturally.
- Decision capture: it records material decisions during the conversation, such as budget, scope, launch timing, and completed modules, rather than storing casual chatter.
- Automatic project creation: when a new topic is discussed for three or more consecutive turns and shows project-like characteristics, it creates projects/<project>/context.md and updates project-registry.json.
Memory is written to the local .workbuddy/context-bridge/ directory by default, with no external upload. If loading context fails, the reply includes a lightweight notice, so users are not left unaware that history is unavailable.

Boundaries

It fits engineering and creator workflows with parallel, long-running projects: marketing plans, product docs, coding tasks, novel chapters, and similar work. It is not an audit-grade solution for highly sensitive data, and it should not replace formal task management or version control. If a project is created by mistake, delete its folder; if a short project name is not recognized, use the full name or update its alias explicitly.