AI skills onboarding checklist for a new teammate

01 / Decision

Choose a handoff that ends in independent use

A useful handoff gives the teammate enough context to make a real choice without doing the task for them. Use one existing recommendation, one representative fixture, and one observable result instead of treating account access or installation as completion.

02 / Workflow

A six-step new teammate handoff

Assume the team library and at least one recommendation already exist. The new teammate should make the important discovery and usage decisions while an owner supplies the task, boundaries, and review criteria.

1. 01

   Choose one real first task

   Pick a task the teammate will actually encounter soon, with a clear input and reviewable output. Keep the first task narrow, free of sensitive data, and small enough to complete in one session. Define the acceptable result before introducing the skill.

   Output: One task brief with a trigger, safe fixture, expected result, and reviewer.

2. 02

   Prepare one team recommendation

   Choose one skill the team already recommends for that task. Confirm that its description, tags, original source, notes, and available use paths are current. Explain why it was selected without presenting the recommendation as a security review or compatibility guarantee.

   Output: One complete, searchable recommendation tied to the first task.

3. 03

   Invite the teammate and set the boundary

   Invite the teammate to the team library and explain the task, expected result, and human review point. Make the product contract explicit: Skills Board points to the latest source, and the teammate still needs to inspect that source and choose a path that fits their setup.

   Output: Confirmed library access plus a short brief covering purpose, limits, and review.

4. 04

   Ask them to find and inspect the skill

   Give the teammate the task language or expected tag, then let them search the library instead of sending a private deep link. Ask them to open the original source and review SKILL.md, supporting files, required tools, permissions, and data paths before use.

   Output: An independently found recommendation with the relevant source and limits identified.

5. 05

   Let them choose a path and test the skill

   The teammate opens the original source, copies a compatible install command, or downloads the latest files as a ZIP. If their agent supports MCP, they can use it to find the same recommendation and retrieve the command. After setting up the skill outside Skills Board, they run the safe fixture and compare the result with the acceptance criteria.

   Output: One observed access and setup path, plus a pass, partial, or fail result from the fixture.

6. 06

   Record the result and next step

   Capture the path and environment used, what the teammate could complete without help, and where the recommendation or setup was unclear. Keep, revise, or replace the recommendation, then name the owner and the source, agent, or workflow change that should trigger another review.

   Output: A handoff decision with evidence, an owner, and a review trigger.

03 / Record

New teammate AI skills handoff checklist

Use one copy per teammate and first workflow. Record only the context needed to improve the handoff, and keep personal or sensitive information out of the checklist.

First task

The trigger, safe fixture, expected output, and person who will review the result.

Team recommendation

The skill, why the team recommends it, its search tags, source, and known limits.

Access and boundary

Library access, allowed data and tools, human review point, and explicit stop conditions.

Independent discovery

The task language used to search and whether the teammate found the recommendation without a private link.

Use path and fixture

The source, compatible command, or ZIP path chosen; whether MCP helped find it; and the observed fixture result.

Outcome and owner

Keep, revise, or replace, with gaps, next action, owner, and review trigger.

# New teammate AI skills handoff

Owner: [person or team]
Teammate role: [role, not personal data]
Handoff date: [date]

## 1. First task
- Trigger: [when this task starts]
- Safe fixture: [representative, non-sensitive input]
- Expected output: [observable result]
- Reviewer: [person or role]

## 2. Team recommendation
- Skill: [name]
- Why the team recommends it: [specific task fit]
- Search terms or tags: [terms the teammate should try]
- Original source: [canonical URL]
- Known limits: [what is untested or unsupported]

## 3. Access and boundary
- Team library access confirmed: [yes/no]
- Allowed tools and data: [scope]
- Human review point: [decision before use]
- Stop when: [permission, data, or requirement is unclear]
- Reminder: a saved skill is a team recommendation, not a security review or compatibility guarantee.

## 4. Independent discovery and review
- Found without a private link: [yes/no]
- Source files inspected: [SKILL.md, scripts, references, templates]
- Required permissions and data paths understood: [yes/no plus notes]
- Latest-source behavior understood: [yes/no]

## 5. Use path and fixture
- Path chosen: [source / compatible command / ZIP]
- Found through MCP: [yes/no/not applicable]
- Agent and environment: [observed setup]
- Acceptance criteria: [pass conditions]
- Result: [pass / partial / fail plus evidence]

## 6. Outcome and follow-up
- What worked without help: [evidence]
- What was unclear: [gap]
- Decision: [keep / revise / replace]
- Next action and owner: [action]
- Review again when: [source, agent, permissions, or workflow changes]

04 / Pitfalls

What weakens a new teammate handoff

05 / Checklist

Ready to share with the team?

• The handoff starts with one real task, safe fixture, expected result, and reviewer.
• One team recommendation has current tags, notes, source, use paths, and known limits.
• The teammate has library access and understands the data, tool, and human review boundaries.
• The teammate found the recommendation by task or tag and inspected its original source.
• The teammate chose a suitable use path and ran the fixture in their actual environment.
• The team recorded the result, remaining gap, decision, owner, and review trigger.