Download the plugin, drop it into Cowork or Claude Code, and go. Eleven talent-acquisition agents turn one role brief into a full eight-document hiring kit.
by Onepromptman · © 2026
Talent One runs in Claude Code and Cowork (the Claude desktop app).
It does not run in the regular Claude chat window. Chat cannot run the kit's eleven specialists, so
set up talent onewill do nothing there. If nothing happens when you try it, you're in the wrong surface.
Option A — marketplace (recommended, two lines):
/plugin marketplace add onepromptman/talent-one
/plugin install talent-one@onepromptman
Option B — download the build: grab dist/talent-one-1.0.3.plugin (or the file attached to the latest release) and add it as a plugin in Claude Code or Cowork.
Then, once:
set up talent one
Five questions about your company, every one with a default and a skip. About five minutes.
Then just ask, in plain words:
| You say | You get |
|---|---|
Run Talent One for a Senior ML Engineer |
The full eight-document kit — research and QA included, ~30 min |
Write me a JD for a staff accountant |
One document, minutes |
Update the comp band to $180k–$220k |
Re-rendered in about a second |
| # | Artifact | Agent | Sample |
|---|---|---|---|
| 01 | Role educational brief (for non-technical recruiters) | Sensei | view |
| 02 | Job description with annotation rail | JD-Bot | view |
| 03 | Talent intelligence map — supply, demand, geo, comp | Atlas | view |
| 04 | Sourcing playbook with tiered boolean library | Hunter | view |
| 05 | 8-touch multi-channel outreach campaign | Shakespeare | view |
| 06 | Interview plan with graded rubrics | Interview Lab | view |
| 07 | Hiring-manager calibration brief | Calibrate | view |
| 08 | 30-minute recruiter screen script | Recruiter Screen | view |
| — | QA audit of the whole package | QA Gate | — |
Relay (plans the run) and Scout (builds the shared per-role research pack) run automatically behind every artifact above. You never invoke them directly.
Every artifact is a polished, self-contained HTML document — zero network requests, opens anywhere. Every number is tagged with where it came from: Sourced, Estimate, or Internal. Nothing is fabricated; gaps are labeled, never filled with filler.
Eight real documents from one Propulsion Engineer run (a fictional company, public labor-market data). Click any preview to open the full interactive HTML.
 01 · Role educational brief — Sensei |
 02 · Job description + annotation rail — JD-Bot |
 03 · Talent intelligence map — Atlas |
 04 · Sourcing playbook — Hunter |
 05 · Multi-channel outreach campaign — Shakespeare |
 06 · Interview plan + graded rubrics — Interview Lab |
 07 · Hiring-manager calibration brief — Calibrate |
 08 · Recruiter screen script — Recruiter Screen |
Previews are screenshots. Each links to the full self-contained HTML — open it in a browser for the live document (GitHub shows
.htmlas source, so use the raw/download view or clone the repo).
Downloads talent-one-1.0.3.plugin · then in Cowork or Claude Code: add it as a plugin and run set up talent one
One research pass per role builds a Context Pack (a shared research summary for the role): market classification, one canonical comp range, requirements, keywords, personas, channels, targets, and timing, each value tagged with where it came from.
Then every requested artifact agent runs in parallel against that pack, producing content only. The finished documents are assembled from fixed templates (not generated by a model), so the formatting is always consistent. A QA gate checks the content against the pack before anything is called done.
role brief ──▶ Relay (plan) ──▶ Scout (one research pass ──▶ Context Pack)
│
┌─────────────────────┼─────────────────────┐
▼ ▼ ▼
Sensei · JD-Bot · Atlas · Hunter · Shakespeare ·
Interview Lab · Calibrate · Recruiter Screen (parallel)
└─────────────────────┬─────────────────────┘
▼
deterministic renderer ──▶ QA Gate ──▶ 8 HTML documents
Timing: full kit ≈ 30 minutes cold (measured live). One artifact on a warm role: minutes. Changing a number later re-renders in seconds.
Anything you already hold — a JD, intake notes, comp data, a brain dump — shortens the pack build. A paste-ready deep-research prompt (skills/talent-one/references/PACK_RESEARCH_PROMPT.md) lets you build the pack in any Claude chat and hand it in as a file.
Works with zero connections, using public data: BLS, O*NET, Census, DOL disclosure data, levels.fyi, live web search. That is a fully supported path, not a degraded mode.
Connecting your ATS or document store upgrades specific agents — see CONNECTORS.md.
Constraints (work authorization, clearance, licensure) are recruiter-written, rendered verbatim, and never invented. An empty constraints list ships no eligibility gate at all.
The plugin is plain text: prompts, templates, and two small scripts that only run inside Claude's sandbox. No executables, nothing installs on your computer, nothing reports back to anyone. The whole source is in this repo — read it.
Verify your download is the official build:
sha256sum talent-one-1.0.3.plugin
# ca7f3c207ab89cce313b6c5c47b92aaeee45794e402aeac74eb2823e934ab88b.claude-plugin/ plugin.json + marketplace.json
agents/ the 11 agent definitions
skills/ 11 skills — the full kit plus one per artifact
samples/ all 8 documents from the built-in worked example
docs/ quickstart PDF, one-pager PDF, release notes
dist/ packaged .plugin build
- Quickstart (PDF) — 3 pages: what it is, how it works, setup
- One-pager (PDF)
- Install guide — per-surface steps for Claude Code and Cowork
- Release notes 1.0.3 · Changelog
- Connectors
Source-available, not open source. You may install it, run it for your own org or your clients, customize it, and use or publish every artifact it generates — those are yours. You may not resell or redistribute the kit itself. Full terms in LICENSE.md.
Artifacts are drafts for professional review. Nothing the kit produces is legal advice; compensation, eligibility, and compliance language must be reviewed by the people responsible for those decisions at your organization.
Questions, bugs, or want a kit run on a live role? Open an issue, or find Bryan (Onepromptman) on LinkedIn.