---
title: "Inspect Gemini CLI context."
description: "Write GEMINI.md guidance, inspect loaded context and distinguish editing from reloading."
canonical: https://editmd.app/learn/lessons/gemini-md/
markdown: https://editmd.app/learn/lessons/gemini-md.md
status: pre-release, waiting list open
generated: editmd-website
---

# Inspect Gemini CLI context.

Lesson 11 of 12 in Markdown School. Write GEMINI.md guidance, inspect loaded context and distinguish editing from reloading.

## A Markdown file can supply tool-specific context

Gemini CLI uses `GEMINI.md` files for persistent instructions. Its documented hierarchy includes global, workspace and more local context. The resulting instructions depend on the working location and configured discovery behavior. [Google's GEMINI.md guide](https://geminicli.com/docs/cli/gemini-md/).

Do not copy a Codex precedence diagram and relabel it Gemini. Similar-looking files are interpreted by different hosts. Start with the smallest project example you can inspect.

## Write an original project contract

```markdown
# Garden project

## Context

Garden notes live in docs/.

## Checks

Run npm test and report failures.

## Boundaries

Review only. Ask before changing files.
```

The command is a supplied fact for our fictional project, not a command every Markdown folder supports. For your own project, check its scripts before including it. “Review only” describes the requested task; host permissions remain the mechanism that controls file writes.

## Inspect before and after an edit

Inside Gemini CLI, these commands serve different purposes:

```text
/memory show
/memory reload
```

`show` displays the combined loaded context. `reload` rescans the configured locations. Edit the disposable project's file, reload, then show it again and look for the exact changed sentence. [Google's context commands](https://geminicli.com/docs/cli/gemini-md/#manage-context-with-the-memory-command).

The browser exercise below does not run those commands. It helps you write the file before testing it in your own installed host.

## Make a useful observation

Try a review request for a short fictional document. Record whether the output reports concerns without editing the file. Compare the file before and after; do not use the agent's own assurance as your only check. Try a folder with additional context only after the single-file test is understandable.

If old guidance remains, inspect loaded sources and configuration. If the context is correct but behavior is wrong, improve the instruction and repeat the small test. These are different diagnoses.

## Project facts versus reusable work

Keep always-relevant project facts here. Put an on-demand procedure in a skill when it has a clear trigger, input and output. Google recommends focused descriptions and loading detailed references only when needed. [Gemini skill best practices](https://geminicli.com/docs/cli/skills-best-practices/).

In the exercise, write the three requested sections. Passing means the browser found that structure and the sample command; it makes no claim about actual Gemini behavior or factual correctness.

Sources checked 17 September 2026. Host behavior can change; verify it against the linked documentation when using these examples.

## Exercise

Write a title and populated Context, Checks and Boundaries sections. Include npm test and an explicit review-only boundary.

## Takeaway

Inspect the context your tool loaded, rather than assuming a filename was enough.

Related guide: https://editmd.app/docs/ai-use/. All lessons: https://editmd.app/learn.md