wishket-scoutlisted

# wishket-scout — Wishket Project Scout & Analysis ## Flow ```mermaid flowchart LR A[Call scan_new MCP] --> B{Any new projects?} B -- No --> Z[Return summary] B -- Yes --> C[Select top N candidates] C --> D[Sequential get_project] D --> E[wishket-analyst dispatch] E --> F[Save to reports/ + Chat summary] ``` ## Step 1: Scan Call `scan_new` on the `wishket` MCP server. Default parameters (`category=development`, `form_factors=web,pc,android,ios`, `max_pages=3`) apply automatically unless specified by the user. - If the user specified keywords, include `keyword`. - On initial run (`baseline: true`), all fetched projects are new; indicate "Baseline Scan" in the report. - If `new_count == 0`: return "No new projects" along with the last scan time (`last_scan` in `~/.wishket-radar/state.db` (SQLite, WAL)) and exit. - If `total_matching_filter` significantly exceeds retrieved items (30 items limit), note in the report: "Fetched top N pages only". ## Step 2: Candidate Selection From the response `new` array, sorted by `match.score` descending: - Select candidates with **score >= 40** or the top 5, whichever is larger. - Include projects with explicit keywords in the title (e.g., Rust, Flutter, LLM) even if the score is lower. - Limit deep analysis to at most 5 items. List remaining items in a brief table at the end of the report. ## Step 3: Detailed Analysis For each candidate, fetch details (including full JSON-LD description) **sequentially** usin