loomlisted

# Loom Workflow Loom is a persistent symbol index. Every session gets faster as agent summaries accumulate. Zero LLM cost — all data from tree-sitter + your own stored summaries. --- ## Session Start Protocol ### Step 1 — Register session ``` start_session(agent_id="claude-code") → { session_id, agent_id, started_at, unannotated_reads, annotation_gaps } ``` Save `session_id` — pass to `get_delta()` at start of **next** session. `unannotated_reads` lists nodes you visited last session but never annotated — prioritize `store_understanding` on these. `annotation_gaps` lists high-value nodes with no summary — good targets for documentation. ### Step 2 — Check delta (if you have a prior session_id) ``` get_delta(previous_session_id="<id>") ``` Returns only changed/deleted nodes since last session. Three response shapes: - **Normal:** `{changed: [context_packets...], deleted: [ids...]}` → review each - **Too many changes:** `{too_many_changes: true, top_changed_paths: [...]}` → treat as fresh start, use `search_code` - **Error:** `{error: "session_not_found"}` → session expired, skip delta, proceed to primer If no prior session_id, pass `agent_id="claude-code"` instead — loom finds your most recent session. ### Step 3 — Load primer (always) Read `loom://primer` resource → ~200-token codebase overview. **Act on primer data:** | Primer Signal | Agent Action | |---------------|--------------| | `empty: true` | Repo not indexed. Tell user: `loom analyze .` — or wait, aut