← ClaudeAtlas

lead-triagelisted

Grade inbound messages (DMs, comments, emails, forms) into A, B, C or SPAM with reasons and a next action, using transparent keyword rules in English and Chinese. Use when an agent watches an inbox and must decide what the owner sees first.
Tommyapi/ai-employee-skills · ★ 0 · AI & Automation · score 70
Install: claude install-skill Tommyapi/ai-employee-skills
# Lead triage Most inbound messages are noise; a few are money. Triage must be cheap, explainable and the same every time, so this skill is rules, not a model call. ## Steps 1. Single message: `python scripts/lead_triage.py --text "Need a quote for 500 units by next week"` 2. Batch: `python scripts/lead_triage.py --jsonl inbox.jsonl` (lines like `{"id": "1", "text": "..."}`) and read the JSON lines. 3. Act on the `action` field: - **A** (score >= 5): reply within an hour with a concrete next step, notify the owner. - **B** (>= 2): reply today with one qualifying question (budget, volume, deadline). - **C**: short polite reply or none; never notify the owner. - **SPAM**: do not reply; archive. 4. Tune: copy `DEFAULT_RULES` from the script into `rules.json`, change words and weights for your market, pass `--rules rules.json`. ## Signals scored budget/price words (3), volume (2), urgency (2), contact intent (2), fit (1), explicit quantity (+1). Any spam phrase wins over all signals. ## Notes - Rules are substring matches on lower-cased text; add your language's words rather than relying on a model to guess. - Review misgrades weekly and add the missed words: that loop is the whole maintenance cost. - Combine with `inbox-ledger` so a graded message is only handled once.