M.A.I. Consulting
Level 7 of seven · free page

Level 7: At the terminal

Run agentic work on folders of files from the terminal, with Claude Code, without writing code: the same thing done to a hundred documents in one supervised run.

2 daysdelivery days from level 6
Task-dependentplanning assumption
EUR 1,600this step, per group of up to nine
Who this level is for

Three people who need exactly this, and nothing more.

NGO and international-organisation staff

A year of reports to review the same way, three hundred documents to restructure, a monthly report built from the same folder of inputs every time.

Employee in a small organisation

Twelve years of procedures, certificates and records that need one consistent, indexed archive before the audit, and nobody to do it.

Student or recent graduate

Two hundred PDFs to rename, classify and summarise into one matrix before the final chapter.

Understand

The background, kept to the minimum sufficient.

Free knowledge, written for a non-technical reader. If you can explain each of these to a colleague in a sentence, you have the background this level needs.

Do

The skills.

  • Batch workRename, restructure, classify or review hundreds of files in one run, with a dry run first that changes nothing.
  • Run the same review across a year of reportsAnd get one consolidated result with a reference to every file and page.
  • Build one repeatable pipelineFor example the monthly donor report from a folder of inputs, with a dry run and a log, and a runbook a colleague can follow.
  • Keep secrets out of filesNo passwords, tokens or keys in the folder or in CLAUDE.md. Use hooks and permission modes to limit what the agent can touch.
Tools

What you need, and only that.

Claude Code (or an equivalent terminal agent), a terminal, git, a scheduled task, and the MCP servers the organisation has approved. A copy of the real folder, never the only copy.
Proof

How we know you are there.

Proof of level 7. One pipeline that runs on real files, with a dry run, a log, and a second person who has run it from the runbook.
Risks that enter here

New at this level, and taught with their refusal cases.

  • An agent that deletes or overwrites, because it was allowed to and nobody asked for a dry run.
  • Credentials in files, where every run and every commit can see them.
  • A pipeline nobody else can run, which is a single point of failure with a nicer name.
  • Work that should have been software: a database, an interface, an integration. Hand it to an engineer.
Skip for now

What the free content pushes at you that you do not need yet.

  • Learning a programming language. It is not needed for this level, and it is a different course.

Not useless, just not yet. Each of these appears at the level where it earns its place, or in a different course.

Cheat sheet

Level 7 on one page.

Print it, pin it next to the screen, and hand it to the colleague who asks how you do it.

Level 7 · At the terminalM.A.I. Consulting · mai4consulting.com/use-cases/level-7/
ItemRule
The terminalA text window: type what to do and where. Claude Code lives there and works on the folder you are in.
Working folderOne folder per job. Copy the files in. Never run on the only copy.
CLAUDE.mdPlain text, in the folder: what the files are, the rules, the output format, what never to touch.
Dry run first"List what you would change. Change nothing."
Permission modesAsk before every write until you trust the run. Never "allow all" on real files.
HooksA rule that runs before an action and can block it: no deletes, no writes outside the folder.
git routineInitialise once. Commit before the run, commit after. Three commands, nothing lost.
Secrets outNo passwords, tokens or keys in the folder or in CLAUDE.md.
LogEvery run writes what it did to a log file. Keep it.
Scheduled runsSame job, same folder, once a month. A person reads the log.
Second runnerSomeone else runs it from the runbook before it counts.
Still engineeringA database, a public interface, other people's credentials, an integration that can fail silently. Send it to an engineer.
Example Claude Code tasks

Realistic, for the work this audience actually does.

Square brackets are yours to fill. Every example keeps personal data out and keeps the decision with the person.

A dry-run rename and classification

In this folder are 312 PDF and DOCX reports from 2019 to 2026. Dry run: for each file, propose a new name in the format YYYY-MM_programme_type_title and the sub-folder it belongs in, by programme. Write the proposal to proposal.csv. Change nothing.

Why it works: The first run on a real folder changes nothing and produces a list a human can read.

The same review across a year of reports

Apply the review checklist in CLAUDE.md to every donor report in reports/. For each, write a one-page review to reviews/ with the checklist items scored and the page reference for each finding. Then write summary.md: the five most common findings across all reports. Do not modify the reports.

Why it works: One method, applied identically to every file, with the originals untouched and the output consolidated.

The monthly pipeline

Build the monthly donor report from inputs/2026-10/ using template.md and the rules in CLAUDE.md. Write to output/2026-10_report.md and a log to logs/2026-10.txt listing every input file used and every figure with its source file and cell. Stop and ask if any input file is missing or if a figure has no source.

Why it works: Inputs, template, output and log are all named, and the stop condition keeps judgement with you.

Consolidation across an archive

Extract the indicator table from each of the 48 quarterly reports in archive/ into one indicators.csv with the columns year, quarter, country, indicator, value, source file, page. Where a value cannot be found, write NA. Do not estimate.

Why it works: A folder-level extraction with a reference on every row and no estimation allowed.

A personal-data sweep before anything else

Scan every file in this folder for personal-data patterns: names in case stories, phone numbers, ID numbers, email addresses. Do not change any file. Write flagged.csv listing file, page and pattern type. I will review each one.

Why it works: The data rule from level 3, applied to a whole archive before any other run touches it.

Guard rails set up and explained

Set up this folder so that you ask before every write, nothing outside this folder can be modified, and no file is ever deleted, only moved to _trash/. Then show me the settings you created and explain each line in plain language.

Why it works: Hooks and permission modes are created and explained, so the person who runs it understands what protects them.

The hand-over runbook

Write RUNBOOK.md for a colleague who has never used a terminal: how to open the terminal in this folder, how to run the monthly report with a dry run first, how to read the log, what to do if it stops, and how to commit before and after. I will ask them to run it while I watch.

Why it works: The proof of level 7 written down: a second person runs it from the guide.

What the training looks like

2 delivery days, each with an output you keep.

Online or in person, in groups of up to nine, on your own tasks and documents. Every day ends with something that did not exist that morning.

Day 1

The terminal, the folder, the rules

What a terminal is; opening it; Claude Code in a copy of one of your real folders. The CLAUDE.md written for that folder. Permission modes and a hook that blocks deletes and writes outside the folder. A dry run on a real batch task (rename, classify or review), then the real run. The git routine before and after.

OutputOne batch task run on a copy of your files, with a dry run, a log and the before-and-after commits.
Day 2

One pipeline, a second runner

Building one repeatable pipeline, for example the monthly report from a folder of inputs, with its log and a scheduled run. The secrets sweep. The "still engineering" test applied to your own list of ideas. A colleague runs the pipeline from the runbook while you watch.

OutputThe pipeline, its runbook, a log from a second person's run, and your level-7 playbook entry.
Price

2 days at EUR 800 per delivery day.

EUR 1,600this step · per group of up to nine
  • Same rate for everyone. EUR 800 per delivery day, whether you come as an individual, a representative sent by your organisation, or a team. Cumulative from level 2: 10 delivery days, EUR 8,000.
  • Shared among the participants. The group price is divided among those who attend; nine people share it nine ways.
  • Online adds nothing. In person adds venue, hotel, hospitality and travel at cost, itemised on the invoice.
  • Time saved: Task-dependent. Level 7 pays off through volume: batch work can exceed any chat figure, and no single number would be honest.

Is level 7 the right next step for you?

Thirty minutes is enough to place you on the ladder and say honestly whether this is the level to book.