Workshop exercise · Level 3, Hour 1 (Intro to Cowork) · ~20 min
A hands-on Claude Code exercise for CPI search consultants. Participants get a folder containing four scattered data files — call notes, a mandate list, a candidate tracker, and client feedback — plus a pre-built interactive "search desk" dashboard that joins them. Their job: ask Claude to change the dashboard so it's designed the way they like it, then save that as a skill they can reuse on fresh data later.
This is the recruiting-tailored version of Every's "Build a Custom Dashboard Skill" exercise. The point isn't the dashboard — it's the muscle: iterate on a real artifact with Claude, then lock your taste into a skill so it's repeatable.
Folder: 00-search-desk-dashboard/
Open search-desk-dashboard.html in a
browser, then follow the step-by-step prompts in
prompt.md: ask Claude to make a change to the
dashboard in the folder, and save that as a skill to reuse for later.
A few things participants might change:
- Design/Brand — colors, typography, dark mode, "make it look like ours"
- Sections/Structure — reorder panels, collapse metrics, add a new view
- Writing Style — labels and action items in their own voice
What's in the folder:
| File | What it is |
|---|---|
search-desk-dashboard.html |
The pre-built dashboard — the artifact they'll customize |
active-searches.csv |
8 live mandates — client, role, strategy, stage, fee, target close, status |
candidates.csv |
13 candidates in play, mapped to searches, with stage + last-contact + interest |
candidate-call-notes.md |
Messy first-person screening notes, one per candidate |
client-feedback.json |
Client interview feedback and ratings on submitted candidates |
All data is synthetic — no real PII. The files cross-reference each other — a call note flags a candidate that the tracker says is stale, that the client feedback says is a "Strong Yes" — and the dashboard is what those joins look like when they're pulled together.
Folder: 01-candidate-submittals/
Two synthetic resumes (Priya Nadkarni, Owen Brackett — the same candidates that later star in the dashboard data) plus step-by-step prompts in prompt.md. Attendees draft a submittal blurb, tune the format once, save it as a candidate-submittal skill, and rerun it on the second resume in a fresh chat. This replaces building the skill on ASCII art — same arc, real artifact.
"Alex" — Search Consultant / VP at CPI. Runs a desk of ~8 live mandates across PE, credit, real estate, hedge fund, VC, and family office. Already fluent with ChatGPT and custom GPTs for processing call notes. Comfortable with AI, brand new to cowork (Claude Code reading local files and building real artifacts).
Pain point the exercise hits: the morning scramble to reconstruct "what needs me today" from four disconnected sources — and the searches that quietly slip because nothing pulls it together.
- Complexity: Easy. The dashboard already exists, so nobody stalls at a blank page — the whole 20 minutes goes to iteration and the skill save.
- Support level: Self-directed. This cohort is AI-forward and will move quickly. Your job is to push them past the first tweak into changes that reflect their taste.
- The mindset shift to name out loud: In their GPTs, they paste text in and get text out. Here, Claude edits a working tool that lives in their files — and a skill makes that edit repeatable on next week's data. That's cowork. Say it explicitly.
- The wow moment: rerunning the
search-desk-dashboardskill in a fresh chat and watching their design come back untouched — same brand, same sections, same voice. If someone's version looks genuinely theirs, have them show the room. - Where they'll get stuck: a few will ask what the "right" change is. There isn't one — Design/Brand, Sections/Structure, and Writing Style are starting points, their judgment fills the rest. That ambiguity is the lesson.
- If they finish early: have them edit a row in
active-searches.csvand rerun the skill to see the dashboard update; stack a second and third change into the skill; or ask Claude to rebuild it automatically every Monday. The automation ask is the bridge to the rest of Level 3. - Data confidence: persona and all data are invented/representative, calibrated to CPI's public business profile (investment-professional search for PE/hedge/credit/RE/VC/family office) and a call-note-heavy workflow. No real candidates, clients, or comp.
Built with Every Consulting's exercise generator.