How-to

Use DM context in a Claude Code task

See Claude Code call dm.prepare_task_memory, apply approved context to a coding task, and report the tested result.

2 min 34 sFor: Developer using Claude Code4 steps

Read the steps OR watch the video. Both teach the same outcome, and neither is needed for the other.Jump to the video

Quick steps

  1. Start from a failing testThe project's tests fail first. Two of them expect a currency column that does not exist yet.
    1. Show the task: cat TASK.md.
    2. Run the tests: python3 -m unittest discover -s tests. Two of the three tests fail.

    What it means: The project's tests fail first. Two of them expect a currency column that does not exist yet.

  2. Ask Claude Code to call DM firstStart Claude Code and ask it to call DM's prepare task memory tool first, with the project, the repository and the kind of work. The operator allowed only a few tools in advance: the DM tools, file edits in this folder, and the test command.
    1. Start Claude Code in the project: claude.
    2. Ask it to call dm.prepare_task_memory with the project, repository, environment, task type and agent type, using the task text.
    3. Allow only what the task needs: the DM tools, file edits in the folder and the test command.

    What it means: Start Claude Code and ask it to call DM's prepare task memory tool first, with the project, the repository and the kind of work. The operator allowed only a few tools in advance: the DM tools, file edits in this folder, and the test command.

  3. Review what DM returnedPress Control O to open the transcript, then scroll up to the DM call. DM returned an approved Decision to use Postgres sixteen on Aurora, a Rule that storage changes must include a rollback plan, and a Skill with the approved procedure. Each item carries its source and the reason it applies. Claude Code decides what to do with them. Nothing enforces them.
    1. Press Ctrl+O to open the transcript, then scroll up to the dm.prepare_task_memory call.
    2. Read the Decision, the Rule and the Skill that came back, each with its source and why it applies.

    What it means: Press Control O to open the transcript, then scroll up to the DM call. DM returned an approved Decision to use Postgres sixteen on Aurora, a Rule that storage changes must include a rollback plan, and a Skill with the approved procedure. Each item carries its source and the reason it applies. Claude Code decides what to do with them. Nothing enforces them.

  4. Check the resultBack in the shell, the same tests now pass. The change is a new migration and its rollback. Claude Code wrote the files. DM supplied the approved context. The tests prove the schema change, not that every rule was followed.
    1. Run the tests again: python3 -m unittest discover -s tests. They pass.
    2. Review the new migration and its rollback with git status and cat.

    What it means: Back in the shell, the same tests now pass. The change is a new migration and its rollback. Claude Code wrote the files. DM supplied the approved context. The tests prove the schema change, not that every rule was followed.

A supervised recording of Claude Code (clip clip-1807f980d4e0, approved by Layout Dryrun). The footage shows only what the operator did in a demo account; it is not a DM step and does not show a native DM integration.

What it means: The task is a ledger storage change: record the currency of each ledger entry amount.

A supervised recording of Claude Code (clip clip-18b4d2ddd8b9, approved by Layout Dryrun). The footage shows only what the operator did in a demo account; it is not a DM step and does not show a native DM integration.

What it means: Claude Code reads the task and calls DM. The call needed no extra approval because it was allowed in advance. DM answers with approved items for this task. Claude Code then reads the project, writes a migration with its rollback, and runs the tests.

A supervised recording of Claude Code (clip clip-6fea26de0a01, approved by Layout Dryrun). The footage shows only what the operator did in a demo account; it is not a DM step and does not show a native DM integration.

What it means: It then reports what it changed, and what it could not check: the tests run on SQLite, not on Aurora.

DM supplies approved context with its sources. The agent still does the work, and people stay in charge.

Descriptive transcript

Read the full transcript, including what happens on screen

Recorded Claude Code session · local demo DM · synthetic data. A real Claude Code session · local demo DM · synthetic data
Contains a recorded Claude Code terminal session against a local demo DM with synthetic data. It is not a published or approved video.

A banner reading "Recorded Claude Code session · local demo DM · synthetic data" stays on screen for the whole video.

00:00–00:10 · intro

On screen: The AuzzurA DM logo, the title "How to Use DM Context in a Claude Code Task" and the subtitle "A real Claude Code session · local demo DM · synthetic data".

Narrator: This tutorial shows how Claude Code uses approved context from AuzzurA DM in a real coding task, in a demo workspace with synthetic data.

00:10–00:19 · task

On screen: A recording of a real Claude Code terminal session (clip clip-1807f980d4e0, approved for editing by Layout Dryrun) plays for 7.6 s. A banner reads "Recorded Claude Code session · local demo DM with synthetic data · private review build". Account details and OAuth URLs are masked at capture time.

Narrator: The task is a ledger storage change: record the currency of each ledger entry amount.

00:19–00:28 · red-tests

On screen: A recording of a real Claude Code terminal session (clip clip-dd5559bcf8c8, approved for editing by Layout Dryrun) plays for 8.0 s. A banner reads "Recorded Claude Code session · local demo DM with synthetic data · private review build". Account details and OAuth URLs are masked at capture time.

Narrator: The project's tests fail first. Two of them expect a currency column that does not exist yet.

00:28–00:47 · prompt

On screen: A recording of a real Claude Code terminal session (clip clip-5521fe26db78, approved for editing by Layout Dryrun) plays for 17.8 s. A banner reads "Recorded Claude Code session · local demo DM with synthetic data · private review build". Account details and OAuth URLs are masked at capture time.

Narrator: Start Claude Code and ask it to call DM's prepare task memory tool first, with the project, the repository and the kind of work.

Narrator: The operator allowed only a few tools in advance: the DM tools, file edits in this folder, and the test command.

00:47–01:27 · work

On screen: A recording of a real Claude Code terminal session (clip clip-18b4d2ddd8b9, approved for editing by Layout Dryrun) plays for 38.6 s. A banner reads "Recorded Claude Code session · local demo DM with synthetic data · private review build". Account details and OAuth URLs are masked at capture time.

Narrator: Claude Code reads the task and calls DM. The call needed no extra approval because it was allowed in advance.

Narrator: DM answers with approved items for this task. Claude Code then reads the project, writes a migration with its rollback, and runs the tests.

01:27–01:37 · answer

On screen: A recording of a real Claude Code terminal session (clip clip-6fea26de0a01, approved for editing by Layout Dryrun) plays for 9.0 s. A banner reads "Recorded Claude Code session · local demo DM with synthetic data · private review build". Account details and OAuth URLs are masked at capture time.

Narrator: It then reports what it changed, and what it could not check: the tests run on SQLite, not on Aurora.

01:37–02:03 · dm-response

On screen: A recording of a real Claude Code terminal session (clip clip-c191ac7ed51f, approved for editing by Layout Dryrun) plays for 25.0 s. A banner reads "Recorded Claude Code session · local demo DM with synthetic data · private review build". Account details and OAuth URLs are masked at capture time.

Narrator: Press Control O to open the transcript, then scroll up to the DM call.

Narrator: DM returned an approved Decision to use Postgres sixteen on Aurora, a Rule that storage changes must include a rollback plan, and a Skill with the approved procedure.

Narrator: Each item carries its source and the reason it applies. Claude Code decides what to do with them. Nothing enforces them.

02:03–02:26 · result

On screen: A recording of a real Claude Code terminal session (clip clip-e0eb1aaf4dce, approved for editing by Layout Dryrun) plays for 21.8 s. A banner reads "Recorded Claude Code session · local demo DM with synthetic data · private review build". Account details and OAuth URLs are masked at capture time.

Narrator: Back in the shell, the same tests now pass. The change is a new migration and its rollback.

Narrator: Claude Code wrote the files. DM supplied the approved context. The tests prove the schema change, not that every rule was followed.

02:26–02:34 · outro

On screen: The AuzzurA DM logo, the heading "Approved context, with its sources" and the line "The agent decides. People stay in charge.".

Narrator: DM supplies approved context with its sources. The agent still does the work, and people stay in charge.

Last verified against the product on .

How-to 2 min 51 s12 steps

Connect DM to Claude Code with OAuth

Connect DM to Claude with OAuth and verify access to DM tools in Claude.ai and Claude Code for approved team context.

For: Developer using Claude Code