/lead-triage

Inbound leads have a half-life. The standard is a first response inside 5 minutes in business hours (knowledge/business-method.md §10.3); this skill exists to clear whatever missed that window, fastest-first — classify each, score by fit, draft the right first response, and tee them up for your approval.

Step 1 — Pull the leads

Run the fetch script. It returns every new lead in the last N hours (default 48), enriched with:

• Source (form submission, inbound email, inbound SMS, inbound call, manual entry)
• Whether they have an opportunity yet
• Contact details (email, phone, company)
• Initial message / form payload
• A fit score (0-100) based on completeness, source quality, and message length
• Recency

python ~/.claude/bos-run.py lead-triage
python ~/.claude/bos-run.py lead-triage --hours 24    # tighter window

Fallback if the script can't run (auth/network): say so briefly, then pull the same window by hand — mcp__trustpager__list_form_submissions, list_email_threads, list_sms_conversations, list_phone_call_logs — and classify from those. Proceed with what you have.

If TrustPager isn't connected at all: say so plainly, then offer the keyless path — the owner pastes the enquiries (form contents, email text) into chat and you classify, score, and draft responses exactly the same way; the drafts land in chat for them to send from their own inbox.

Step 2 — Classify each lead

Age matters as much as fit: a fresh engaged lead outranks an older high-score one, and anything unworked ~3 months is cold-list territory, not triage (§10.3, directional). Engaged leads — the ones who replied, wrote a detailed message, or called — are the currency; a raw name-and-email handle is not one yet.

For each lead, propose a category:

Show the classification + reasoning to the user. The user can override per lead.

Step 3 — Draft the per-lead response

After classification is agreed, draft the message for each lead. The draft should:

• Use the lead's first name (never "Hi there", never "Hi friend")
• Reference something concrete from their enquiry (their stated need, their company name, their question) — this is the moat against generic templates
• End with a single clear next step ("Are you free Wed or Thurs morning for a 15-min call?", not "Looking forward to hearing from you!") — a concrete time, never "let me know"
• Match the tone of the workspace — read recent sent emails to calibrate before drafting

Step 4 — Send with approval (per lead)

For each draft:

• Show the user: lead name, category, channel, the proposed message.
• Wait for explicit yes/no per lead. NEVER batch-send.
• On yes: send via mcp__trustpager__send_email or send_sms for the per-lead channel.
• On no: ask what to change, or skip.

Important behaviours

• Create the opportunity first if missing. Every fast-track lead becomes an opportunity in your pipeline before the message goes out. The skill should call create_opportunity with the right pipeline + stage + contact link.
• Disqualify ≠ delete. Mark with a "lost" status + reason so you can review later. Never delete.
• A genuine-fit 'no' is not lost. If a lead replies not-now or too-expensive, never offer the same thing cheaper; offer a changed scope or payment shape (business-method.md §9.4) or park them in nurture with a dated next touch.
• Spam heuristics. Free email + generic body + "SEO services / web design proposal / partnership opportunity" = disqualify by default.
• Don't promise specifics you don't know. "Our standard package starts at $X" — only if the user told you the figure or it's in workspace knowledge.
• Quiet hours. Same as missed-call-recovery — before 7am / after 8pm = email not SMS.

Output shape

End with a one-line summary: "Triaged N leads — F fast track sent, S nurture sent, C cold sent, D disqualified, R skipped for your review."