← AI Leadership Lab: Reports, Skills, Agents and Business Planning
Block VIII · Step-by-step tutorial

Set up Code and source access

Run the local calculator, install your Code Skill and inspect source permissions.

0 of 7 steps complete

Saved in this browser only. Sign in with an active enrolment to save new progress to your account. Step checks are separate from marking the lesson complete.

Outcome and preparation

Day 2 · 09:15–10:30 · 75 minutes, after a 09:00–09:15 recap. Open your canvas, approved local practice folder and Claude Desktop. Node.js 22+ should be available from prework. Allocate 10 minutes to orientation, 50 to setup and 15 to checks.

Claude Code can read files, run commands and edit a working folder. Review its proposed actions. Start with the dedicated fictional pack, not your home directory or a production repository.

Step by step

  1. In Claude Desktop open Code, choose a local session and select the unzipped ai-leadership-lab folder. Inspect any folder-trust prompt before accepting. Confirm the displayed folder matches your practice copy. Keep the normal permission mode; do not enable blanket bypass permissions.

  2. Ask Code to inspect README.md, CLAUDE.md and the source register, then list inputs without modifying anything. Open the files it cites to confirm the folder boundary.

    Prompt / example
    We are working only in this fictional workshop folder.
    Read README.md, CLAUDE.md and reference/source-register.md.
    List approved data files, reporting periods and permitted output locations.
    Do not read outside this folder, change inputs, install anything or contact external systems.
    Explain any proposed command before running it.
    
  3. Open Terminal/PowerShell in this folder and run the following, or ask Code to run the exact commands after you inspect them. node --version should show 22 or later. Save actual command output to your rehearsal log.

    Prompt / example
    node --version
    node report.mjs --period 2026-10-05 --previous 2026-09-28 --as-of 2026-10-12
    
  4. Open outputs/2026-10-05/report.md, calculations.json and run.json. Check the period, source digests and draft status. Run the same command again: expect status: unchanged, the same output path and no second report. Do not edit the generated files; save human notes in review.md.

  5. Ask Code to adapt yesterday's Skill to the local workflow and save it at .claude/skills/executive-report/SKILL.md. Review the diff before accepting. Compare with completed/.claude/skills/executive-report/SKILL.md after your attempt. Invoke /executive-report with explicit period and as-of date.

    Prompt / example
    Adapt my Skill to this folder and save .claude/skills/executive-report/SKILL.md.
    Use report.mjs for arithmetic, outputs/ for drafts, and reference/ for definitions.
    Ask for missing dates. Never relabel the historical data as current.
    Preserve generated reports and stop on changed inputs. Do not send or publish.
    Show me the file and how to invoke it before running anything.
    [Paste the Skill from Day 1]
    
  6. Practise source access through one already approved read-only connector. In App settings/Customize, open Connectors, select the approved provider and inspect the displayed account and requested scope. If it requests broader access than approved, stop and ask IT. Read a named public/fictional document, record its locator, then ask the same domain question answered from the local file. Never paste a database password into chat.

  7. Save handover/source-access.md: actual folder, permitted tools, connector/account if used, granted scope, source locators and revocation steps. If none is approved, record local-only; connector demo observed, not integrated and proceed with the local MVP.

Expected output and verification

A working local Code session, deterministic report, unchanged rerun and project-scoped Skill. A connector exercise is evidence of integration only if you actually read the intended source through it. MCP provides tool access; it does not guarantee correct results or authorise all data access.

Troubleshooting

  • Node not found: install the official Node LTS and reopen your terminal/Code session.
  • ENOENT: confirm the pack was completely unzipped; the runner resolves inputs beside itself.
  • Permission denied: inspect the exact requested operation and allow only what your workshop folder requires.
  • Skill unavailable: check the path/name and reopen the session. Follow current Skill documentation.
  • Connector blocked: keep the local workflow and record the gap; do not install an unapproved MCP server.

Completion checklist

Prompt examples can be copied with the Copy button. Tick the tutorial steps as you work through them.