What problem does it solve? Memory cascades only flow down one ancestor chain, so lessons filed in a sibling project or a separate knowledge tree never reach the current project on their own. This Skill performs the inbound gather: it finds knowledge stored elsewhere that is relevant to the current project and imports it safely, without dangling cross-tree references or leaked credentials. ## Core Features & Use Cases - Three-stage gather pipeline: a cheap deterministic keyword grep (gather_scan.py) finds candidates grouped by knowledge tree, a context-isolated subagent inspects them for genuine relevance, and imports follow strict tree rules (lift to a common ancestor, copy locally, or labeled copy across trees). - Privacy scrubbing: secrets and PII are stripped before anything crosses a project or tree boundary, while concrete operational detail is preserved. - Debounce tracking: a (project, topic) record via --mark/--seen flags prevents re-gathering the same topic on every trigger. - Use Case: When seeding a brand-new project, run the gather with the project's scope descriptor to pull in relevant lessons from your other projects so the new project starts informed instead of blank. ## Quick Start Ask the AI to collect knowledge relevant to this project from your other projects and memory trees, for example by saying "collect knowledge about deployment lessons from my other projects".
