dpd-find-similarlisted

# /dpd-find-similar — Retrieval-augmented proposal (v0.3.2) Announce: "Using dpd-find-similar skill." This skill is the **user-pull entry point** for DPD's retrieval-augmented proposal pipeline (v0.3.2 spec §6). ## When to invoke User invokes `/dpd-find-similar [query?]`. Claude may NOT auto-fire this skill (v0.3.2 §3.4: user-pull only). ## Step 1 — Query resolution If `query` argument is present, use it directly. Otherwise, propose a query candidate in this priority and explicitly confirm before running the search: 1. **focus_node text** (call `get_session_state` to fetch). 2. **Recent conversation topic** (cluster the last few turns around the most prominent anchor). 3. **Active Pool item cluster** (call `pool_list(active_only=True)` and look for a shared keyword among items whose attach point is undetermined). Wording: "query: `<candidate>` で検索する、OK?" — Wait for explicit confirmation. Do not silently proceed. ## Step 2 — Run `find_similar` ```text find_similar(query="<confirmed>", top_k=5, include_open=False) → { "results": [SubgraphSummary, ...] } ``` `SubgraphSummary` fields: `start_node_id`, `session_id`, `root_id`, `scope`, `start_text`, `end_text`, `achievement_conditions`, `state` (`closed` / `archived`), `score`, `matched_snippet`, `closed_at`. ## Step 3 — Present results Render in conversation prose (not a graph diagram). Format: ```text 過去類似 subgraph (state=closed/archived): 1. [score=8.21] scope=<scope> root=<root_id> start=<start_node_id> "<sta