How-to

Prepare relevant context for a task

Create a bounded context package from approved knowledge. Before your AI agent changes code, give it the decisions that apply.

2 min 8 sFor: Coding-agent user10 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. Click Prepare contextBefore an AI agent starts work, you can check which approved knowledge DM would prepare for it. That way you see the decisions, rules and skills that apply before any work begins, not after. Open Prepare context from the sidebar.
    1. Start on DM Home (/).
    2. Click "Prepare context". The next view is "Prepare context" (/task).

    What it means: Before an AI agent starts work, you can check which approved knowledge DM would prepare for it. That way you see the decisions, rules and skills that apply before any work begins, not after. Open Prepare context from the sidebar.

  2. Fill in the formDescribe the work in your own words. There is no special syntax. Here, the work is a migration that changes the invoices table.
    1. Type "Add a migration that changes the invoices table" into the "What are you working on? Required" field.

    What it means: Describe the work in your own words. There is no special syntax. Here, the work is a migration that changes the invoices table.

  3. Click Advanced context OptionalThe advanced context is optional. It lets you add the project, repository, environment and who will do the work. The more you add, the better DM can tell which policies apply. We'll leave it empty for now.
    1. Click "Advanced context Optional". The next view is "Prepare context" (/task).

    What it means: The advanced context is optional. It lets you add the project, repository, environment and who will do the work. The more you add, the better DM can tell which policies apply. We'll leave it empty for now.

  4. Click Prepare contextThen select Prepare context.
    1. Click "Prepare context". The next view is "Prepare context" (/task).

    What it means: Then select Prepare context.

  5. Review what was foundDM selects only approved knowledge that applies to this work. The summary counts the items it found, the missing context, and any conflicts.
    1. Review "What was found": "3 Relevant knowledge items 0 Missing context 0 Conflicts More context needed Approval con…".

    What it means: DM selects only approved knowledge that applies to this work. The summary counts the items it found, the missing context, and any conflicts.

  6. Review how it was preparedThe provenance card shows how DM prepared this. Here, it matched on keywords, because semantic retrieval isn't configured in this workspace.
    1. Review "How it was prepared": "Selection provenance How DM prepared this authorized context. Retrieval keyword · depende…".

    What it means: The provenance card shows how DM prepared this. Here, it matched on keywords, because semantic retrieval isn't configured in this workspace.

  7. Review an approved decisionHere it found an approved decision. Production schema changes need two reviewers. Each item says why it applies.
    1. Review "An approved decision": "DECISION active Any migration that touches production schema needs approval from two revi…".

    What it means: Here it found an approved decision. Production schema changes need two reviewers. Each item says why it applies.

  8. Review a ruleIt also found a rule and a skill that follow from that decision. The rule says what must happen. The skill describes how to do the work.
    1. Review "A rule": "RULE Review production migrations twice Production schema migrations MUST be approved by…".
    2. Review "A skill": "SKILL Prepare a safe schema migration Steps an agent follows to draft a migration that me…".

    What it means: It also found a rule and a skill that follow from that decision. The rule says what must happen. The skill describes how to do the work.

  9. Review approval contextThe approval card is honest about what it could not determine. Without more detail about the actor and the project, DM can't tell whether a policy applies. It says more context is needed.
    1. Review "Approval context": "Approval context More context needed Current authorization context No active authorizatio…".

    What it means: The approval card is honest about what it could not determine. Without more detail about the actor and the project, DM can't tell whether a policy applies. It says more context is needed.

  10. Review conflictsThe conflicts card would flag knowledge that disagrees with other current knowledge. None was returned here. This page shows what DM would prepare. It doesn't send anything to an agent, and DM can't guarantee that an external agent follows the rules it is given. Next, connect the agent to use DM tools.
    1. Review "Conflicts": "Conflicts No related conflict signals were returned.".

    What it means: The conflicts card would flag knowledge that disagrees with other current knowledge. None was returned here. This page shows what DM would prepare. It doesn't send anything to an agent, and DM can't guarantee that an external agent follows the rules it is given. Next, connect the agent to use DM tools.

The decision, rule and skill are invented for this tour. A real workspace returns its own approved knowledge that applies to your task and that you are allowed to see; selection, ranking and wording in a live workspace were not shown or verified.

Advanced context fields (project, repository, environment, agent), policy matching, repository evidence and episodic memory are not demonstrated. Handing the context to an agent through MCP tools is a separate episode.

Next lesson

Next: Create an agent and set up Claude Code

Descriptive transcript

Read the full transcript, including what happens on screen

Demo environment · synthetic workspace and data · backend responses are simulated. See which approved knowledge DM would hand an agent
This guide tours the real DM Prepare context page with a synthetic workspace and a simulated backend response. DM prepares context from approved knowledge you can access; it does not guarantee that an external agent follows it.

Disclosure (kept in this transcript and the video description, not drawn on the frames): Demo environment · synthetic workspace and data · backend responses are simulated.

00:00–00:17 · start

On screen: Start on DM Home (/). Click "Prepare context". The next view is "Prepare context" (/task). A highlight marks "Prepare context".

Narrator: Before an AI agent starts work, you can check which approved knowledge DM would prepare for it.

Narrator: That way you see the decisions, rules and skills that apply before any work begins, not after.

Narrator: Open Prepare context from the sidebar.

00:17–00:27 · type

On screen: Type "Add a migration that changes the invoices table" into the "What are you working on? Required" field. A highlight marks "The work".

Narrator: Describe the work in your own words. There is no special syntax.

Narrator: Here, the work is a migration that changes the invoices table.

00:27–00:43 · advanced

On screen: Click "Advanced context Optional". The next view is "Prepare context" (/task). A highlight labelled "Optional details" marks "Advanced context Optional".

Narrator: The advanced context is optional. It lets you add the project, repository, environment and who will do the work.

Narrator: The more you add, the better DM can tell which policies apply. We'll leave it empty for now.

00:43–00:47 · submit

On screen: Click "Prepare context". The next view is "Prepare context" (/task). A highlight marks "Prepare context".

Narrator: Then select Prepare context.

00:47–01:00 · summary

On screen: A highlight labelled "What was found" marks "3 Relevant knowledge items 0 Missing context 0 Conflicts More context…".

Narrator: DM selects only approved knowledge that applies to this work.

Narrator: The summary counts the items it found, the missing context, and any conflicts.

01:00–01:11 · provenance

On screen: A highlight labelled "How it was prepared" marks "Selection provenance How DM prepared this authorized context. Retriev…".

Narrator: The provenance card shows how DM prepared this.

Narrator: Here, it matched on keywords, because semantic retrieval isn't configured in this workspace.

01:11–01:21 · decision

On screen: A highlight labelled "An approved decision" marks "DECISION active Any migration that touches production schema needs ap…".

Narrator: Here it found an approved decision. Production schema changes need two reviewers.

Narrator: Each item says why it applies.

01:21–01:31 · rule

On screen: A highlight labelled "A rule" marks "RULE Review production migrations twice Production schema migrations…". A highlight labelled "A skill" marks "SKILL Prepare a safe schema migration Steps an agent follows to draft…".

Narrator: It also found a rule and a skill that follow from that decision.

Narrator: The rule says what must happen. The skill describes how to do the work.

01:31–01:46 · approval

On screen: A highlight labelled "Approval context" marks "Approval context More context needed Current authorization context No…".

Narrator: The approval card is honest about what it could not determine.

Narrator: Without more detail about the actor and the project, DM can't tell whether a policy applies. It says more context is needed.

01:46–02:07 · conflicts

On screen: A highlight labelled "Conflicts" marks "Conflicts No related conflict signals were returned.".

Narrator: The conflicts card would flag knowledge that disagrees with other current knowledge. None was returned here.

Narrator: This page shows what DM would prepare. It doesn't send anything to an agent, and DM can't guarantee that an external agent follows the rules it is given.

Narrator: Next, connect the agent to use DM tools.

Last verified against the product on .

How-to 1 min 54 s10 steps

Find why a decision was made

Retrieve rationale, evidence and change history without chasing old messages. You found the implementation. But why was that choice made?

For: New engineer or tech lead

How-to 1 min 20 s9 steps

Review and approve a decision candidate

Demonstrate why AI extraction is not automatically team policy. DM proposes what it found. Your team decides what becomes trusted.

For: Reviewer / workspace decision owner

How-to 49 s7 steps

Add your first evidence to DM

Start from one relevant source rather than recording everything. A team decision often starts in a document, a ticket, or a meeting note.

For: Workspace member