Talent One

For recruiters and talent teams opening a role: describe the role once and get nine ready-to-use hiring documents, from the job description to the interview plan.

by OnePromptMan

_{Made from a fictional search. Your documents arrive as HTML files you open in any browser, offline.}

Agent pack (Claude plugin) · No coding · Live · MIT

What you get

• Nine hiring documents for one role. A role brief for you, the job description, a talent map, a sourcing playbook, outreach, an interview plan, a hiring-manager brief, a screen script and an interviewer briefing.
• Facts you can check. Every market figure says where it came from, and a separate checker challenges every draft before you see it.
• Drafts only. It never sends outreach, posts a job or writes to your ATS. You decide what to share.

Get started

1. Install. In Claude Code, paste these two lines:

   /plugin marketplace add onepromptman/Talent-One
   /plugin install talent-one@talent-one
   

   In Cowork (the Claude desktop app), upload the plugin instead.
2. Set up. Type set up talent one. Five questions about your company, each with a default and a skip. About five minutes, less if another kit already ran setup: the profile is shared. Per-surface steps, uninstall, and upgrading from Talent One 1.x: docs/INSTALL.md.
3. Run. Ask Run Talent One for a Senior ML Engineer. It asks only what it cannot look up, then works for about half an hour. Open the nine HTML documents in your working folder.

For one document, ask Write me a JD for a staff accountant. Update the comp band to $180k-$220k re-renders in about a second.

What you need: Claude Code or Cowork, and Python 3.10 or later where the plugin runs. No API key and nothing else to install. It does not run in the regular Claude or ChatGPT chat window, because only Claude Code and Cowork can start its helper agents.

Upload the plugin in Cowork

1. Download this repository (Code, then Download ZIP) and unzip it.

2. Open the plugin folder, the one whose .claude-plugin folder holds plugin.json, and zip its contents so that .claude-plugin/plugin.json sits at the top of the zip. On macOS or Linux, from inside that folder:

   zip -r ../talent-one.zip .

   In Windows PowerShell, from inside that folder:

   Compress-Archive -Path .\* -DestinationPath ..\talent-one.zip

3. In the Claude desktop app open Customize, then Plugins, then Add, then Upload plugin, and choose the zip.

Setup is a conversation

Setup confirms your company and its website, looks up what it can, then asks what you hire, any rule every candidate must meet, how the documents should look and whether to connect a research tool or your ATS. Every question except the company has a default and a skip. Connectors sign in through your browser; setup never asks for a key and never asks you to edit a file. Your answers are saved in onepromptman-profile.md in your working folder, which the other OnePromptMan kits share.

See it before you install

Download this repository as a ZIP, unzip it, and open any file in samples/ in your browser, for example samples/02_Job_Description_SAMPLE.html. They are all nine documents from a fictional search and need no account.

Role brief for the recruiter	Job description, with notes on each choice	Talent map: where the people are
Sourcing playbook	Outreach: email, LinkedIn, phone	Interview plan with scoring guides
Hiring-manager brief	Screen script for the first call	Interviewer briefing and debrief form

The fictional first-run checks show what a passing run should produce.

How it works

flowchart LR
  A["You describe<br/>the role"] --> B["It asks what it<br/>cannot look up"]
  B --> C["One research pass<br/>builds the role pack"]
  C --> D["Nine specialists draft<br/>at the same time"]
  D --> E["A checker challenges<br/>every draft"]
  E --> F["Fixed templates<br/>format each document"]
  F --> G["A final check of the<br/>whole package"]
  G --> H["Nine HTML documents<br/>in your folder"]

Loading

A full nine-document run takes about half an hour and a couple of dozen model calls on your plan; one document takes minutes. The call-by-call breakdown is in docs/how-it-works.md.

What it asks of your data

• What you type and the research it runs pass through your Claude session (Anthropic). Connectors get only what an agent asks for, after you sign in.
• The documents are about the role, not about candidates. If you paste candidate or employee details, they are personal data: share only what the role needs.
• The profile, research pack and documents stay in your working folder until you delete them.
• Optional connectors: Exa or Bright Data (web research), Greenhouse or Ashby (your ATS). Nothing is sent to one until you sign in. See CONNECTORS.md.

If something goes wrong

• set up talent one does nothing. You are in the regular chat window. Use Claude Code or Cowork.
• The first render stops and says Python is missing or too old. Install Python 3.10 or later from python.org, then ask again.
• The final check returns a document. Its finding names the agent and the fix. Ask for the fix and that agent patches its draft.

Run it without opening a chat

A headless runner (runner/) runs the same pipeline from a terminal for a scripted or scheduled run, with the claude CLI signed in. It is for people comfortable with a terminal; see runner/README.md.

For engineers

Settings

Setting	Default	What it does
onepromptman-profile.md	written by setup	Company, roles, constraints and brand, shared by every kit
Document prefix	TK	Code on each document (TK-01 to TK-09); set it in setup's Brand step
Footer credit	on	"Built with Talent One by OnePromptMan"; setup can turn it off
.mcp.json	four connectors, signed out	Exa, Bright Data, Greenhouse, Ashby
LOG_LEVEL, LOG_FORMAT, LOG_FILE	WARNING, text, standard error	What the scripts log, and where

Troubleshooting

Scripts log to standard error, or to LOG_FILE; LOG_LEVEL=DEBUG shows each step. The renderer's one package, beautifulsoup4, ships inside the plugin in skills/talent-one/scripts/_vendor/ (version, licence and the two small edits are in its README), so nothing is installed at run time.

Build and test

This repository is generated from the talent-one module of onepromptman/talentai by that repository's tools/publish_talent_one.py, and is never edited by hand: change the module there and publish again. Issues are welcome here.

python3 -m venv .venv && . .venv/bin/activate
python -m pip install -r requirements-dev.txt ruff==0.17.3
python -m pytest -q

bash evals/run_gate.sh      # acceptance gate: briefs, context budget, lint, render + verify x9, hygiene
bash build-plugin.sh        # dist/talent-one-<version>.plugin, from committed files only

On Windows, run the gate from Git Bash with PYTHONUTF8=1 set, so the verifier's JSON is written as UTF-8.

The version is in .claude-plugin/plugin.json and MODULE.yaml; changes are in CHANGELOG.md. Agents, model tiers, layout and the sample gallery: docs/how-it-works.md. The profile file's shape is in PROFILE-SPEC.md.

License

MIT, Copyright (c) 2026 OnePromptMan. Every document the kit generates is yours. They are drafts for professional review, not legal advice: compensation, eligibility and compliance language must be checked by the people responsible for them. Questions or bugs? Open an issue.