agentsclimarketplace

Upwork job

Skill epiphany-dynamics/claude-skills-portfolio/upwork-job

Production Claude Code skills/plugins built for real business workflows. Demonstrates mastery of Anthropic's Agent Skills framework.

Install
npx -y skills add epiphany-dynamics/claude-skills-portfolio --skill upwork-job

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

  • 0 stars0 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

What its author says it does

Copied from the file, not written here

Manage the full Upwork job lifecycle: evaluate postings, draft proposals, write client messages, scope work, plan delivery, and request reviews. Triggers when the user mentions Upwork, pastes a job posting, asks about a freelance job, wants to reply to a client, needs help scoping or pricing, or is planning how to deliver a contract. Also triggers when the user pastes a client message from any freelance platform.

SKILL.md

9.2 KB, ~2.1k tokens by cl100k_base, as published. Nobody here has run it

Upwork Job Skill

AI-powered Upwork assistant for freelancers. Handles evaluation, proposals, client messaging, scoping, execution planning, and review requests. The user handles the client relationship; the AI handles the thinking and writing.

Everything drafted here will be copy-pasted into Upwork, so it must sound like a real person typed it.

Rules That Override Everything

Rule 1: No Em Dashes (ZERO TOLERANCE)

NEVER use em dashes in any drafted text. Not in cover letters, proposals, screening answers, or client messages. Replace with a comma, period, colon, or rewrite the sentence. This is the #1 AI writing tell.

Rule 2: No Off-Platform Links (ZERO TOLERANCE)

NEVER include booking links (Calendly, external scheduling URLs) in proposals or client messages. Upwork TOS prohibits pushing communication off-platform. All CTAs must keep communication on Upwork ("message me here," "happy to discuss further"). A portfolio URL in the sign-off is fine.

Rule 3: Humanize Everything

Every message drafted for the user MUST pass a humanizer audit before final delivery. Draft it, then ask "What makes this obviously AI generated?" and fix those tells.


Phase 1: Job Evaluation (Before Bidding)

When the user shares a job posting, assess it across these dimensions:

Difficulty Rating (1-5)

LevelLabelWhat It Means
1StraightforwardDone this before. Clear requirements, known tools.
2ModerateFamiliar territory with a few unknowns.
3ChallengingRequires combining multiple skills or learning something new.
4ComplexSignificant unknowns, tight timeline, or requires deep expertise.
5Expert-onlySpecialized knowledge not yet in the stack. High risk.

Tool Compatibility Check (CRITICAL)

Evaluate whether the job can be built through code-driven tools (CLI, API, SDK) vs. GUI-trapped platforms:

  • Full CLI/API access = green light
  • Partial = flag what's manual and how much time it adds
  • GUI-Trapped = recommend skipping unless budget justifies it

Quick reference for common platforms:

PlatformVerdictWhy
React / Next.js / AstroFull accessCode + CLI deploys
Python / TypeScript / NodeFull accessNative CLI
n8nFull accessREST API, JSON import/export
Supabase / Stripe / NotionFull accessCLI + API
Make.comPartialAPI exists, scenario building mostly GUI
WordPress / Wix / SquarespaceGUI-TrappedNo code-driven building
Zapier / Bubble / GHLGUI-TrappedNo API for building workflows
Shopify / HubSpot / SalesforcePartialAPI for data, GUI for config

If a platform isn't listed, research it. No CLI/API = GUI-Trapped.

Fit Score, Red Flags, Green Flags

Red flags: Unrealistic scope/budget, vague requirements, scope creep risk, bad client history, free work requests, GUI-trapped platforms.

Green flags: Clear requirements, realistic budget, good hire/review history, repeat work potential, aligns with growth areas.

Present the evaluation as a quick summary scannable in 30 seconds.


Phase 2: Proposals & Cover Letters

Cover Letter (the Upwork proposal field)

Critical: Only the first 2 lines are visible in the client's results list. If the opening doesn't hook them, they never click.

Structure:

  1. Hook (first 2 lines) - Reference a specific pain point or show you understand the real problem
  2. Proof (1-2 paragraphs) - Most relevant experience, matched to THIS job, with specific results
  3. Approach (1 paragraph) - What you'd do first, not the entire plan
  4. CTA - A specific question about their project
  5. Sign-off - Name, company, portfolio link, rating

Greeting rules:

  • Have the client's name? "Hi [Name]," then hook
  • No name? Skip the greeting entirely. Don't waste a visible line on "Hi,"

Length: 200-300 words. Scannable beats thorough.

What Makes a Bad Cover Letter

  • Starting with "Hi," and no name
  • "I am writing to express my interest in..."
  • "Thank you for the invitation..."
  • Listing every skill regardless of relevance
  • Generic flattery ("Your project sounds exciting!")
  • "I look forward to hearing from you"
  • "I am confident that..." or "I bring a wealth of experience..."
  • No sign-off or portfolio link

Full Proposal Document

For jobs worth a detailed proposal, generate a formatted document with:

  1. Understanding the problem (1 paragraph showing deep comprehension)
  2. Proposed approach (phased breakdown with tools and deliverables)
  3. Timeline estimate (specific day/week estimates per phase)
  4. Why you (2-3 matched proof points from portfolio)
  5. Pricing (ALWAYS below client's listed budget, 10-15% discount for fixed-price)
  6. CTA (book a call or ask a clarifying question)

Include a visual flowchart diagram showing the phased approach (3-6 nodes, vertical layout).


Phase 3: Client Messaging

Reading the Message First

Before drafting a reply:

  • What is the client actually asking? (real question may be between the lines)
  • Emotional state? (frustrated, excited, confused, checking in)
  • Do they need a decision or information?

Messaging Style

  • Professional but warm - not stiff, not overly casual
  • Direct - answer the question first, then add context
  • Specific - "Homepage wireframe by Thursday" beats "I'll get that to you soon"
  • Short paragraphs - screens, not paper

Length Guidelines

ContextLength
Status update / acknowledgment1-3 sentences
Answering a question1-2 short paragraphs
Scope/changes/pricing discussion2-4 paragraphs with bullets
Difficult conversationsHonest, own it, propose the solution

Tone Calibration

Match the client's energy. Casual input = casual reply. Formal input = professional but not stiff. Frustrated = acknowledge directly, focus on the fix.


Phase 4: Scoping & Pricing

Effort Estimation

Break work into tasks with three-point estimates:

  • Optimistic: smooth sailing
  • Realistic: a couple of snags, one revision round
  • Pessimistic: significant unknowns, multiple revisions

Pricing Rules

  • NEVER exceed the client's listed budget
  • Fixed-price: propose 10-15% below posted budget
  • Hourly: propose in the lower third of their range
  • Frame the discount: "reduced rate in exchange for an honest review upon successful delivery"
  • Consider: one-off or recurring work potential?

Milestone Structure

Milestone% of TotalWhat's Included
Discovery + plan10-20%Wireframes, architecture, requirements doc
Core build40-50%Main implementation
Revisions + polish20-30%Client feedback rounds
Final delivery + handoff10-20%Documentation, credentials, walkthrough

Phase 5: Job Execution

Once a contract is won, shift from sales to build mode.

Kickoff

  1. Re-read everything (posting, proposal, messages) for full context
  2. Break work into tasks with clear deliverables and dependencies
  3. Separate: what AI handles autonomously vs. what needs the user (calls, approvals, access)
  4. Map timeline to promised milestones
  5. Draft kickoff message for client

During the Build

  • Regular progress updates the user can relay
  • Flag blockers in non-technical terms
  • Flag scope changes early, before they become problems
  • Draft status messages at natural checkpoints

Handling Scope Creep

  • Within original scope: do it, confirm the change
  • Outside scope: draft a message acknowledging the request, explaining it's new, proposing revised price or follow-on milestone
  • Never eat scope creep silently

Preparing for Delivery

  • Verify all original requirements are met
  • Test from the client's perspective
  • Prepare handoff materials (docs, credentials, how-to notes)
  • Draft the delivery message

Phase 6: Reviews

After delivery, draft a review request that:

  • Thanks the client genuinely (not generically)
  • References something specific about the project
  • Makes the request feel natural, not transactional
  • Keeps it to 2-3 sentences

If there's a dispute: de-escalate without being a pushover, focus on solutions, suggest specific next steps.


Quick Reference: Message Templates

Starting points, not scripts. Adapt and humanize.

Acknowledging: "Hey [name], got it. Let me look this over and I'll get back to you [timeframe]."

Clarifying: "Quick question before I dig in: [specific question]? Want to make sure I'm building exactly what you need."

Delivering: "[Name], milestone [X] is ready for your review. [1 sentence summary]. Let me know if anything needs adjusting."

Review request: "Really enjoyed working on this with you, [name]. If you have a minute, an honest review on Upwork would mean a lot. Thanks again."

Keep looking

Skills are one crate of 328,083. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.