shapelex-memorylisted

# ShapeLex Memory ## Overview Use ShapeLex proactively as a navigable memory layer, not as lossy reconstruction. ShapeLex is agent-driven by default: decide when it helps, and do not wait for the user to say "use ShapeLex" during long or repeated-context work. Prefer exact source text when the context is short or when compressed output fails to save tokens. ShapeLex primarily saves input tokens by replacing old context with searchable handles. Save output tokens by answering tersely: report actions, results, blockers, and exact next steps; avoid narrative explanations unless the user asks. ## Workflow 1. Briefly tell the user the first time you use ShapeLex in a session, for example: "I am going to compress older context with ShapeLex to keep this session lighter." Do not repeat this for every tool call. 2. Compress only long or repeated context. Use `shapelex_compress_text` for documents/code-like text and `shapelex_compress_messages` for older conversation history. Prefer `sourcePath` over `text` when the source already exists in the workspace. 3. If a compression result has `compressionSkipped: true`, keep using the returned exact `compressedText`; do not force handles into the prompt. 4. Use `shapelex_memory_overview` when starting work, when the user asks what memory/session is active, when switching projects, or when cleanup might be needed. 5. Recommend lean mode for normal work. Suggest full mode only when the user needs lower-level search, retrieve, explain, ris