Upwork Automation Using Ai

Automate Upwork login, search, filter, and proposal drafting in one browser session, stopping before submission unless explicitly instructed.

Security Scan

The skill's requested access and instructions match its stated purpose (browser automation of Upwork job search and proposal drafting) and it does not request unrelated credentials, installs, or persistent privileges.

Purpose & Capability

The name/description describe browser-driven Upwork automation and the SKILL.md contains step-by-step browser automation instructions (login, search, filter, open job, fill proposal). No extraneous environment variables, binaries, or installs are requested — this footprint is consistent with the stated purpose.

Instruction Scope

Instructions explicitly perform interactive browser actions (including login flows, handling captcha/2FA/manual checkpoints, clicking Apply, filling proposal fields) and require reading page text/screenshots. The skill refuses to persist credentials and states it will stop before submission unless told. This scope is appropriate for the task, but it means the agent will handle sensitive credentials provided at runtime and will read page content — users should be aware of that behavior.

Install Mechanism

Instruction-only skill with no install spec and no code files — nothing is downloaded or written to disk by an installer, which minimizes installation risk.

Credentials

No environment variables, primary credential, or config paths are required. The skill relies on user-provided credentials at run time only (per its text), which is proportionate to the stated functionality.

Persistence & Privilege

always is false, no installs, and the SKILL.md forbids persisting credentials to files. The skill does not request persistent system presence or elevated privileges.

Assessment

This skill is coherent with its purpose, but it automates a browser and will ask you to provide Upwork (or Google) credentials interactively for a run. Before using it: (1) only provide credentials when you are ready to perform the session and avoid giving long-term API tokens or password dumps, (2) prefer testing on a non-critical account first, (3) confirm the agent shows screenshots/state checks before any sensitive clicks, and (4) verify it does not submit proposals unless you explicitly say “submit now.” If you rely on Google SSO, expect cross-site authentication flows and possible 2FA prompts — the skill may ask you to complete those manually. If anything looks unexpected (requests to save credentials, to access unrelated services, or to submit without confirmation), stop and revoke access.

Workflow

1. Keep one active browser session/tab unless user explicitly asks to switch.
2. Open Upwork and ensure login state.
3. If logged out, run login automation:
   • go to Upwork login page
   • prefer "Continue with Google" when user asks for Gmail login
   • fill email/password only from user-provided credentials for this run
   • complete required next step (password, captcha/2FA/manual checkpoint)
   • verify logged-in indicators before continuing
4. Collect/confirm criteria:
   • allowed job types/stack
   • minimum client quality thresholds
   • disqualifiers
   • proposal constraints (submit or draft-only)
5. Search jobs and shortlist visible matches.
6. Reject jobs with any disqualifier.
7. Pick the best remaining job (or top N if user asks).
8. Open job detail and click Apply.
9. Detect proposal location:
   • if same tab, continue
   • if new tab/window, switch to it (or ask user to activate once if tool cannot switch)
10. Fill proposal fields completely.
11. Stop before submission unless user explicitly says submit.

Hard Rules

• Do not submit proposal unless user explicitly says to submit now.
• Stay in the same tab/session unless user requests switching.
• After clicking Apply, immediately check whether proposal opened in same tab.
• If proposal opens in a new tab/window, auto-switch to that tab when the tool supports tab targeting.
• If auto-switch is not supported by the active tool, instruct user to activate the new tab once, then continue there.
• If a modal blocks actions, close modal first, then continue.
• Validate on-page state with screenshot/text checks before risky clicks.
• Never persist credentials to skill files or notes; use credentials only for the active run.

Job Filtering Rubric

Treat a job as valid only if all required checks pass.

Required:

• Job type matches user scope (ecommerce stack or general development)
• Posted within time window (default <= 3 days unless user changes)
• Client quality meets thresholds (e.g., avg hourly >= $10 if visible)
• Fixed budget >= $100 when fixed-price budget is visible

Disqualify when any is true:

• Individual-only hiring restriction
• Urgent/start-today pressure language
• No/poor payment history when user disallows it
• Requires screen share, onsite reporting, strict time tracking, or skill tests (if disallowed)

If uncertain from visible data:

• mark as "needs manual review"
• do not claim it fully passed

Proposal Drafting Template

Use concise, specific structure:

1. Direct fit opening (stack + outcome)
2. Delivery plan (milestones/timeline)
3. Relevant proof (similar builds)
4. Communication cadence
5. Clear CTA

Prefer concrete numbers and short bullets over long paragraphs.

Execution Notes (Browser Reliability)

• Prefer deterministic selectors and verify each transition with screenshot/get_text.
• If generic selectors misfire, target by nearby unique text.
• If automation cannot switch to newly opened tab, ask user to bring proposal tab active and confirm.
• Keep browser open at end and report exact completion status:
  • job selected
  • proposal fields filled
  • submission state (not submitted)

Output Format to User

After completing work, respond with:

• Selected job: <title>
• Why selected: <criteria match summary>
• Fields filled: <key fields>
• Submission: Not submitted
• Next action: “Review and tell me ‘submit now’ if you want me to send it.”