Upwork Proposal Skill — SkynetLabs Edition

Goal: produce proposal client wants to reply to. Not CV-dump. Not generic. Not "I hope this finds you well."

Inputs

User provides ONE or BOTH of:

1. Upwork JD (full job post text)
2. Client message (their first DM/reply after invite)

If only JD given → write proposal. If only client message given → write reply (skip the JD-extraction step, mirror the message instead). If both → write proposal that anticipates the reply concerns.

If user gives neither, ask: "Paste the JD (or invite message). Anything you already know about the client/budget?"

Workflow

Step 1 — Red-flag scan (skip-or-bid decision)

Before writing, check JD for:

• Payment method not verified → skip if job <$1K
• "Send a sample first" / "do a test task" → free-work scam, skip
• Budget <20% of stated scope ($50 for "build me Uber") → skip
• "Communicate on Telegram/WhatsApp first" → TOS violation + scam pattern, skip
• 0 hires + posted >2 weeks → ghost client, deprioritize
• Spelling/grammar inconsistencies → AI-generated scam, skip
• "NDA before any conversation" → skip
• "Urgent — need today" + no budget listed → panic client, churn risk
• Hire rate <30% with 50+ jobs posted → shopper not buyer
• Vague ("build me an app like Uber") → unscope-able, skip

Output a 1-line verdict before the proposal: Verdict: BID / BID-WITH-CAUTION / SKIP — reason.

If SKIP, stop. Do not draft.

Step 2 — Extract from JD

Pull these 8 signals (mentally, not in output):

1. Stated problem — what they typed
2. Hidden problem — underneath (e.g., "need fast site" = previous dev ghosted)
3. Tech stack constraints — named tools = non-negotiable
4. Decision-maker tells — "we / my team / founder / agency"
5. Budget signal — range, "open", or absent
6. Timeline urgency — "ASAP", "no rush", date-anchored
7. Past-pain phrases — "last freelancer", "tried before", "didn't work"
8. Success metric — revenue / speed / launch / compliance

Reflect 2 of these in the opener. Never all 8 (parrot territory).

Step 3 — Pick the proof point (Profile→Proposal bridge)

From your real builds, weave ONE per plan-bullet. Rotate from this bank:

Niche	Proof point	Live URL / receipt
WordPress	A custom companion plugin, HMAC+replay auth, kills ZIP uploads	your-personal-site.com
n8n	50+ flows shipped, AEO content engine, LinkedIn auto-post, FB clone engine	example.com
WhatsApp bot	A real-estate client, €850 / 5-day MVP	(ask for client perm before naming)
Healthcare site	A colon & rectal surgeon pre-launch	example.vercel.app
Wellness funnel	6-page editorial WP, Upwork wellness brief	wellness-funnel-demo.vercel.app
Clinic site	A UK clinical lead nurse	(private — describe shape, not name)
AEO/SEO	CiteLift / aeo-audit-tool 5-API stack	citelift.app (in build)
Local-services	A home-services NYC quote-funnel	(private — case study OK)
Demo-first habit	Builds free demo before pitch	shows in pattern, not bullet

Rule: ONE proof per bullet, woven, not stacked. Live URL > screenshot > claim. Never list >2 URLs in proposal — clients click 1.

Step 4 — Decide: Loom and/or speculative micro-deliverable

Loom worth it when: job ≥$1K, scope is concrete, client has ≥1 prior hire. 60–90 sec, screen-record on their site/brief, give 3 specific recommendations.

Speculative demo worth it when:

• job ≥$500
• scope concrete (a site, a flow, a funnel)
• client payment-verified
• you can ship demo in <2 hrs

Neither when: vague JD, $50 budgets, "send samples first" requests (free-work scam), 0-hire ghost clients.

If recommended, say so in Meta block (after proposal). Don't actually build the demo until user confirms.

Step 5 — Draft proposal (120–180 words)

Skeleton:

1. Hook (2 lines, ≤25 words line 1) — mirrors job, never starts with "I"
2. 3-bullet plan — each bullet = action + "Done = ..." outcome + woven proof
3. 1 proof artifact — inline link, not "see portfolio at end"
4. 2-3 sharp questions — micro-demo of expertise (see Q-bank below)
5. Price anchor — concrete number + scope + timeline
6. Presumptive CTA — choice-based ("two ways to start: A or B"), signed first name only

Hook templates (pick the fit)

• Mirror + claim: "[Specific JD pain] is almost always [diagnosis] — fixable in [timeline]."
• Diagnostic hook: "Your post mentions [tool A] AND [tool B] — order matters here, and most builds get it backwards."
• Pattern-match flex: "Just shipped a near-identical [thing] last week ([client/context], [scope+price+timeline])."
• Specific compliment: "Your brief is already 80% spec'd — rare. Only gap I see is [X]."

Banned openers: "I hope this finds you well", "I am the perfect fit", "I have X years experience", "Dear Hiring Manager", "I came across your job post".

Question bank (pick 2-3, never lazy)

Lazy (banned): "What's your budget?" "When do you need this?" "What's the goal?"

Sharp:

• "Is the [WhatsApp bot / chat flow] routing to a human after qualification, or fully autonomous?"
• "n8n self-hosted or cloud? (Changes credentials approach.)"
• "Existing GA4 + Search Console access, or do we set up tracking from zero?"
• "Shopify theme — Dawn-based or custom/legacy? (Affects upgrade path.)"
• "When you say 'AI integration' — Claude/GPT API direct, or via Make/n8n wrapper?"
• "WordPress — managed host (WP Engine/Kinsta) or shared (BanaHosting/Hostinger)? Affects deploy method."
• "Is the success metric leads, bookings, or revenue? Build sequencing changes."

Pricing language (anchor, never apologize)

• Anchor: "$2,500 fixed for the scope you described — [scope items], [revisions], [timeline]. Done = [observable artifact]."
• Tiered (high-ticket): "Three ways to start: (A) $400 discovery sprint — audit + spec + clickable wireframe. (B) $1,800 build — implements A. (C) $3,500 build + 30 days post-launch support."
• "What's your budget?" jobs: "Comparable builds I've shipped run $1,200–$3,000 depending on [variable]. Happy to scope to either end once I know [Q1, Q2]."
• Hourly defense: "$65/hr, capped at 25 hrs for this milestone — over is on me."
• Race-to-bottom defense (last resort): "Not the cheapest bid you'll get — the one who's already shipped 3 of these. If lowest price is the brief, no hard feelings."

Closing CTA — presumptive close

Banned: "Looking forward to hearing from you", "Let me know if interested", "Hope to chat soon", "Best regards / SkynetLabs Team."

Alive:

• "Two ways to start: 15-min call this week, or I send a 2-slide plan tomorrow — your pick. — [Your Name]"
• "Want the demo URL? Reply 'yes' and it's in your inbox in an hour."
• "If [Q1] is yes, discovery doc to you tomorrow. — [Your Name]"
• "Worth a 10-min call Thursday? I'll come with the wireframe pre-built."

Step 6 — Output format

Always output in this structure:

═══════════════════════════════════════
VERDICT: <BID / BID-WITH-CAUTION / SKIP>
Reason: <one line>
═══════════════════════════════════════

PROPOSAL (paste-ready):
---
<the proposal — 120-180 words>
---

META:
- Word count: <n>
- Connects: <est cost> | Boost: <YES if 9/10 fit + verified + ≥$1K else NO>
- Loom: <YES + what to show / NO>
- Speculative demo: <YES + what to ship / NO>
- Red flags found: <list or "none">
- Anticipated client objections: <2-3>
- If they reply "what's your timeline / when can you start": <prepared answer>
═══════════════════════════════════════

Reply mode (client already messaged back)

If user pastes a client reply, switch to reply mode:

• Mirror their last message in line 1 (quote a phrase or restate their concern)
• Answer ONE concern fully, not all of them (avoid wall-of-text)
• Always end with ONE forward-motion question
• Length: 60–120 words, plain text, no bullets unless they used bullets
• Match their formality level (formal client → formal you; casual → casual)

Hard rules

• Never start a proposal with "I"
• Never write more than 200 words
• Never list more than 2 URLs
• Never use "Looking forward to hearing from you"
• Never quote a price as a range without the upper-bound logic ("$X–$Y depending on Z")
• Never offer free trial week / first task free / money-back guarantee
• Never lie about a build (no fake metrics, no invented clients) — see feedback-no-fake-claims
• If naming a real client — only first names or shape ("a real-estate developer", "a UK clinical nurse"), never full client identifiers without permission

Quality checklist (run before delivering)

• Opener doesn't start with "I"
• Opener mirrors ≥1 specific JD detail (not just rephrases JD)
• 3 plan bullets, each with "Done = ..." or proof link
• Concrete price stated
• 2-3 sharp questions, no lazy ones
• CTA is presumptive, not "looking forward"
• Word count 120-180
• Signed "— [Your Name]" not "Best, SkynetLabs Team"
• No fake claims / invented metrics
• Loom + spec-demo decisions stated

Connects economics (for Boost decision)

• $0.15/Connect. Standard bid = 6C ($0.90). Competitive = 10–16C.
• Plus plan: $19.99/mo, 100 Connects. Worth past ~50 bids/mo.
• Boost only when ALL true: 9/10 fit, client payment-verified, hire rate >50%, job ≥$1K, <20 proposals already.
• Healthy reply rate: 20–35%. Below 15% = hook or targeting wrong, not volume.

When to invoke

User says any of:

• "write upwork proposal" / "upwork bid" / "draft a proposal for this Upwork job"
• "/upwork-proposal"
• pastes an Upwork JD
• "client just messaged back, what do I say"
• "help me reply to this Upwork client"

When NOT to invoke

• Generic cover letter requests (LinkedIn, email outreach, RFP) — different game
• Existing client comms inside an active Upwork contract — that's not a proposal
• Fiverr gig descriptions — different platform mechanics

---

Reference: Full research synthesis in playbook.md (loaded only if user asks "why" or wants deeper teardown).