Agent reviewing LinkedIn prospecting evidence

LinkedIn prospecting, inside your agent

Turn a target-account brief into evidence you can review

Dux-Soup helps an AI agent organize company context, professional-role clues, and public profile signals without pretending that activity equals intent. You decide what is relevant before any outreach begins.

Open the four-state runbookInspect a sample result
Entity identityDated contextHuman approvalExplicit unknowns

Installation

Copy, run, configure, then inspect

Each state has a different success signal. A copied command is only clipboard content; a completed run is only a functioning package.

  1. 1. Copy

    npx -y @okki-global/okki-go-taroball

    Success: the exact command is in your clipboard. Next: open a controlled terminal.

  2. 2. Run

    Inspect the package prompt, requested permissions, network behavior, and completion output. Stop if the runtime or requested scope is unclear.

  3. 3. Configure

    Store API keys outside prompts and repositories. Grant minimum scope, name every connected source, and define where temporary output is retained.

  4. 4. First result

    Use a small cohort you already know. Check company identity, role relevance, observation dates, exclusions, provenance, and unresolved fields.

Auditable preview

See the claim, its basis, and its review state

SAMPLE

Prospect result

Company: Northstar Systems [illustrative]

Role: Operations leader · relevance pending review

Unknown: current project timing

Data behavior

A ledger instead of a badge wall

Data sources

Config-dependent. Record each connected provider, permitted purpose, field provenance, and observation time. Public profile context can be stale or incomplete.

API keys

Operator-controlled. Keep secrets in the runtime secret store, grant minimum scope, rotate them after testing, and never include them in shared logs.

Runtime behavior

Inspect locally. Review permissions, browser or API activity, retries, rate limits, errors, and destinations before connecting production systems.

Retention and deletion

Config-dependent. Define who can access results, how corrections and suppression propagate, and when temporary records are removed.

FAQ

Before the first live workflow

Does copying install anything?

No. The copy button changes only the clipboard. Installation begins only when you deliberately run the command in your terminal and accept its prompts.

Does a profile signal prove buying intent?

No. Hiring, role changes, posting activity, and company events are research clues. They need a dated source and a human interpretation.

Can the workflow send messages automatically?

Keep research, drafting, approval, and sending as separate gates. Verify relevance, suppression, opt-out handling, applicable regional rules, and sender-domain health first.

What should the first test measure?

Measure false matches, unresolved identities, event freshness, evidence availability, review time, and correction effort on a known cohort—not an unsupported universal accuracy rate.