⏳ This skill is pending AI review.
Scores will appear once the review pipeline completes.
thumbtack-draft
>-
Choose how to use this skill
You do not need every option. Choose the path your AI client supports. The stable page stays the same; versioned files are immutable.
1. Native installer
This listing has no registered native installer command. Use the complete package or source fallback below, depending on what your client supports.
Do not guess an installer command or replace an existing version without reviewing the diff.
2. Complete package recommended
Download the ZIP when available. It includes SKILL.md plus the references, security notes and version metadata.
No complete ProSkills package is published for this listing yet.3. Prompt-only
Copy the prompt above when the agent can read the stable page or when you want to adopt the workflow without installing a skill.
Need only the instruction file?
Download SKILL.md only if your client requires a single file. The complete ZIP is safer for a full installation because it preserves the references and release context.
No path installs or executes anything by itself. Your agent still needs access to the project files. Before updating, compare the installed version and review the diff.
// RATINGS
Not yet listed on ClawHub or SkillsMP
// README
openclaw-thumbtack-draft
A reusable openclaw skill that drafts replies to inbound Thumbtack leads for solo home service pros — designed to win the conversation in the first 60 seconds.
This is the methodology layer behind Novanest Pro, my AI system that runs a real US home services business solo from Barcelona. The production system has live API credentials, real prices, and customer-specific tuning that aren't in this repo. What's in this repo is the structural pattern: how to think about a Thumbtack reply, what wins, and what kills the deal.
If you're a solo service pro, an AI builder working on lead-conversion tools, or just curious how a multi-agent system handles real customer messages — this is the skill spec.
What's in this repo
| File | Purpose |
|---|---|
SKILL.md | The full skill specification: orchestrator pattern, pricing framework structure, the proven reply formula, hard rules, mid-conversation rules |
example_replies.md | Worked examples of the formula applied to common lead types (sanitized — no real customers) |
LICENSE | MIT |
What's NOT in this repo
- Actual prices. The production system uses task-based pricing tuned to Hovo's specific business and California market. The structure of the framework is in
SKILL.md, but the numbers are placeholders. Plug in your own. - Internal API endpoints / tokens. Removed.
- Customer data or specific lead history. Removed.
- The Python backend (FastAPI, SQLite, webhook intake, dispatch pipeline). That's a separate, private codebase. See novanest-pro for the architecture overview.
Why publish this
I built Novanest Pro by analyzing 100+ real Thumbtack lead conversations and reverse-engineering what made some win and others lose. The pattern isn't proprietary — it's verifiable craft. Any pro who reads this skill spec can apply it manually or wire it into their own system.
The thing that's hard to copy is doing it consistently in under 60 seconds for every lead, automatically, while you're on the ladder doing the actual work. That's what the production system handles. The methodology in this skill is what the production system encodes.
Use it as an openclaw skill
If you're running openclaw:
cd ~/.openclaw/workspace/skills/
git clone https://github.com/nova-novanest/openclaw-thumbtack-draft.git thumbtack-draft
The SKILL.md frontmatter will be auto-detected. Tune the prices in the pricing table to match your own market, plug in your base location for the travel-surcharge step, and you're ready to draft.
Use it as a standalone reference
SKILL.md is just markdown. You can read it, copy it into your own playbook, or adapt the principles to a totally different lead source (Yelp, Angi, Google Local Services). The orchestrator pattern, the "match the description length" rule, the "answer the literal question first" rule, the "never offer free on-site estimate" rule — none of those depend on Thumbtack specifically. They're generic conversion patterns.
License
MIT — use this however you want, including commercially. Attribution appreciated but not required.
Built by Hovhannes Hovhannisyan (Hovo) from Barcelona. Originally extracted from Novanest Pro.
// HOW IT'S BUILT
// FILES
2 files shipped with this skill. Inline text is the catalog copy. The upstream link is the source repository.
SKILL.md7.9 KBupstream ↗
--- name: thumbtack-draft description: >- Generates a winning reply to an inbound Thumbtack lead for a solo home service pro. Implements the orchestrator + sub-agent pattern, applies a task-based pricing framework, and uses a verified "winning formula" reply template optimized for under-60-second response times. Use whenever a new Thumbtack lead arrives or when responding to a customer mid-conversation. --- # Thumbtack Draft Generation Skill A reusable openclaw skill for drafting Thumbtack lead replies that actually convert. > **Note on prices:** Every dollar amount in this skill is a **placeholder**. The methodology is the value — plug in your own task-based rates from your own market. --- ## Orchestrator Pattern (mandatory) Your orchestrator (e.g. Claude Opus) is the decider. It does NOT do research itself — it delegates to a sub-agent worker. **Flow:** 1. **Orchestrator reads the lead** — understand the person, what they need, their emotional state. A few tokens, no tool calls yet. 2. **Spawn a research sub-agent** for location/context lookups. Give it a focused task brief: ``` sub_agent.spawn(task=""" Research this Thumbtack lead for a reply draft. - Customer first name: [name] - Location: [city, zip] - Category: [category] - Description: '[verbatim from lead]' Do these lookups and return findings only (no draft): 1. Drive time from [YOUR_BASE_LOCATION] to [zip] 2. Neighborhood profile for [zip] — type (suburban/urban/apartments), median home value 3. Anything notable about the area or customer Keep findings under 150 words. Return facts only. """) ``` 3. **Research returns → orchestrator writes the draft** using the intel + the pricing framework + the winning formula below. 4. **Patch the draft** back to wherever your system stores pending replies (e.g. an approval queue, a Telegram message, a webhook callback). This pattern saves ~80% of orchestrator tokens while keeping draft quality high. **For obvious leads** (e.g. a single doorbell install in a familiar zip), skip the research spawn and draft directly. --- ## Input The skill receives: - `customer_name` — first name only is fine - `location` — city + zip code - `category` — Thumbtack's category label - `description` — the verbatim text the customer typed - `details` — Thumbtack's checkbox/dropdown answers (treat as form noise unless they conflict with the description) - `lead_price` — what the platform charged you for this lead --- ## Step A — Price the Job Build a **task-based** pricing table for your own market. The structure looks like this: | Task | Placeholder Price | |---|---| | Camera setup, normal height (Wi-Fi config only) | `$X` | | Camera setup, ladder needed (15ft+) | `$X + ladder fee` | | Indoor smart camera (wall mount, app pairing) | `$X` | | Outdoor wired camera (cable management, clean run) | `$X` | | Battery camera install (no wiring) | `$X` | | Floodlight camera (electrical to junction box) | `$X` | | Smart doorbell (wired existing chime location) | `$X` | | Smart doorbell (no existing wiring, needs solar or battery) | `$X` | **Pricing rules:** - Add up the tasks, no home-value multiplier. **Task cost is task cost.** - If the customer already owns the hardware (check `details` for the "already have the security features" answer or the equivalent), price labor only — no hardware markup. - If the camera count is unclear (e.g. "1–2 cameras"), give a per-camera price. Don't assume a count. **Why task-based and not flat:** Customers comparison-shop on flat quotes. A per-task breakdown anchors the conversation in *what you're doing*, not *what you're charging*. It also lets you price-discriminate honestly per job without negotiating. --- ## Step B — Travel Surcharge (silent) Run a drive-time lookup from your base location to the customer's zip: - **Under 45 min** → no surcharge - **45–75 min** → add a flat surcharge to your total (suggested: small, e.g. ~$25). **Never mention to the customer.** Bake it into the quote. - **Over 75 min** → larger flat surcharge (suggested: ~$75). Same — silent, baked in. Mentioning a travel surcharge to a Thumbtack customer is a documented conversion killer. They feel charged for something they don't see value in. Bake it into the price and present a single number. --- ## Step C — Hardware Check Check `details` for any field that says the customer already has the equipment. - **They have it** → price labor only - **They don't** → you can either price labor + hardware-at-cost or labor only and ask them to buy specific products **Recommendation:** for installs under $400, price labor only and ask the customer to buy the hardware. It's cleaner, you avoid markup negotiation, and you avoid being responsible for the wrong product. --- ## Step D — Read the Description This is the highest-leverage step. Most pros skip it. - **Description = the only real input.** It's the only thing the customer actually typed with their own hands. - **Details/checkboxes = form noise.** They're filtered through Thumbtack's form UI and don't reflect how the customer actually thinks. Use them only to confirm what the description already implies. - **Count words.** Match the description length in your reply, +/- 30%. If they wrote 8 words, write 10–15. If they wrote 80 words, write 100. Mismatching length is a tone failure customers notice subconsciously. - **Match vocabulary.** If they write casually, you write casually. If they use technical terms, you use technical terms. Mirror them. - **Do NOT repeat what they said back to them.** *"So you need cameras installed?"* is the death sentence of an opener. They know what they wrote. They want to know what you're going to do. --- ## Step E — Write the Reply ### The winning formula ``` Hey [Name]! Just saw your note — sounds like a quick setup. $[price] [installed/all in]. [Today/Tomorrow] works. When's good? ``` This template was extracted from a sample of historical hired-vs-not-hired leads. It's not magic — it's the structural elements that win: 1. **Greeting + acknowledgment** in one line 2. **Price up front** in dollars, not "starting at" 3. **Concrete availability** today or tomorrow, not "this week" 4. **One question, asking WHEN** (not WHAT) ### Variations | Situation | Variation | |---|---| | Customer already owns hardware | *"You've got the system — $[X] installed and tested. What days work?"* | | Need one piece of info to price | Ask for **photo OR count** — never both. Then price in the next message. | | High-stakes / anxious customer | Add one reassurance line: *"installed and tested before I leave."* | | Low lead price (cheap job) | Same formula, lower price, faster turnaround offer. Don't downgrade tone. | ### Hard rules - **"Just saw your note — sounds like a quick setup"** — use this verbatim. It tested better than every variation I tried. - **Give the price in message 1 or 2, always.** *"I'll get you a quote"* is a near-zero conversion phrase. - **Offer today or tomorrow** — never "this week," never "next week." - **One question max — and the question is WHEN, not WHAT.** Asking *"what kind of cameras?"* hands the conversation back to the customer's homework. Asking *"when's good?"* moves toward booking. - **Match description length.** 8 words in → 10–15 words out max. 80 words in → 100 words out. - **Never offer a free on-site estimate.** Customers who accept free on-site estimates are documented to ghost after the visit at much higher rates than customers who book on a phone quote. - **"Installed and tested before I leave"** is a strong value signal. Use it on jobs that justify it. - **Never send two messages before the customer replies.** If your previous outbound has no reply yet, hold. --- ## Mid-Conversation Replies When the customer has already replied to your first message and you need to respond again, the rules shift slightly: - **Answer their LITERAL question first** — never skip paREADME.mdupstream ↗
Contents are not inlined in the catalog. Open the upstream file.
Plain-text copy, no login: /skill-bundles/nova-novanest-openclaw-thumbtack-draft.md