AI Skills for Claude and ChatGPT
Ready-made recipes that teach a connected assistant how to use Stori well: review a shortlist, debrief a candidate, launch a role, write the Monday hiring update, run a careful outbound search. Install one in a minute, edit any line of it.
What a Skill Is
Once Stori is connected to Claude or ChatGPT, the assistant can call every Stori tool your workspace allows. A skill tells it how to use those tools for one job: which to call, in what order, what to show you, and where to stop and ask. It is a short Markdown file. Your taste, your roles and your candidates stay in Stori; the skill only describes the procedure.
Without a skill, you explain the procedure each time. With one, you say "review the Product Designer shortlist" and the assistant pulls the pool, tests the role's requirements against the interview transcripts, quotes each candidate with a timestamp, and asks before it writes a note or moves anyone.
Every skill follows Stori's rules: it says who should be considered, never who is best; it quotes candidates rather than paraphrasing them; it treats a locked interview as identity only; and it confirms before anything that writes, spends a credit or reaches a person outside your workspace.
The Skills
| Skill | What it does | File |
|---|---|---|
| Shortlist review | Pulls a role's pool, tests each requirement against the transcripts, and writes a tiered review with quotes and open questions. Offers notes, stage moves and review requests, and asks first. | SKILL.md · zip |
| Interview debrief | One candidate's interview organised by the role's requirements: evidence, gaps, tensions between answers, and questions for the next round. | SKILL.md · zip |
| Role launch | Turns intake notes into a spec written as things the hire will have done, creates the job as a draft, and explains Teller before anything goes live. | SKILL.md · zip |
| Hiring pulse | The Monday update: roles, pipeline by stage, interviews booked, offers out, Teller proposals and reviews waiting, and where things are stuck. | SKILL.md · zip |
| Source and reach | An outbound search with agreed filters, a shortlist you pick from, contact details revealed one decision at a time, and individually written drafts you read before anything is sent. | SKILL.md · zip |
Candidates have their own skills for interview preparation and outreach. They are listed in the candidate guide.
Install in Claude
- Connect Stori first, following Connect Claude to Stori.
- Download the skill's zip from the table above.
- In claude.ai open Settings → Capabilities → Skills and upload the zip. Claude Desktop uses the same setting. In Claude Code, unzip it into your
~/.claude/skills/folder. - Start a new chat and ask in plain language. Claude picks the skill from your request: "debrief Priya for the Staff Engineer panel" is enough.
Install in ChatGPT
- Connect Stori first, following Connect Any AI Assistant.
- Open the skill's SKILL.md from the table and copy its text.
- Create a Project in ChatGPT, add the Stori connection to it, and paste the text into the project's instructions. A custom GPT's instructions work the same way.
- Ask in plain language inside that project.
Make It Yours
The file is plain Markdown and every line is editable. Teams usually change three things: the shape of the output (a table your hiring managers already read), the house rules (always private notes, never move anyone past the second stage without a named reviewer), and the example requests, so the assistant recognises the phrases your team actually uses. Keep the operating principles; they are what make the output defensible.
A skill can only use tools your workspace has switched on in Settings → Integrations → AI Apps, and only within the permissions the person who connected approved. Sourcing, revealing contact details and sending outreach are off for new workspaces until an admin turns them on. A skill that reaches a switched-off tool is refused and told which switch to ask about.
What Skills Cannot Do
- Read another company's data, or a candidate whose interview is locked to your workspace.
- Rank people or name a best candidate. The tools do not return that, and the skills do not pretend to.
- Send an offer document. That stays a deliberate act in the app.
- Act without a signed-in member having connected the assistant and, for anything external, having confirmed in the chat.
Every change a skill makes is in your audit log against the person whose connection made it, the same as a change made in the app.