Skip to main content
Agents don’t just analyze—they take action. From creating pull requests and posting comments to sending Slack messages, agents can perform a wide range of operations based on what they discover. This guide covers how agents take actions and send notifications. Actions Diagram

​
How Actions Work

Agents accomplish actions through tools. You describe what you want done in your prompt, and the agent uses the appropriate tools to make it happen. For example: 
prompt: |
  If documentation is missing:
  1. Generate the documentation
  2. Create a new branch
  3. Commit the changes
  4. Open a pull request
  5. Post a summary to #data-team on Slack
The agent will use:
  • Write / Edit to create/update files
  • Bash with git commands to commit and push
  • Bash with gh pr create to open a PR
  • slack_tool to send the Slack message

​
Git Actions

Agents perform Git operations through the Bash tool using standard Git and GitHub CLI commands.

​
Creating Commits

prompt: |
  After updating the documentation files:
  1. Stage the changes: git add models/
  2. Commit with a descriptive message: git commit -m "docs: Update model documentation"

​
Creating Branches

prompt: |
  Create a new branch for your changes:
  git checkout -b docs/update-customers-model

​
Pushing Changes

prompt: |
  Push your changes to the remote:
  git push -u origin docs/update-customers-model

​
Creating Pull Requests

Use the GitHub CLI (gh) to create pull requests: 
prompt: |
  Create a pull request with your changes:
  gh pr create \
    --title "docs: Update customers model documentation" \
    --body "This PR updates documentation based on current data profiles.

  Changes:
  - Added column descriptions
  - Updated row count estimates
  - Added null rate annotations

  Auto-generated by Buster."

​
Posting PR Comments

prompt: |
  Post a comment on the current PR summarizing your analysis:
  gh pr comment $PR_NUMBER --body "## Analysis Complete
  
  Found 3 models with missing documentation.
  
  See the created PR for fixes."

​
Auto-Commit Behavior

For pull request triggers, the auto_commit field controls commit behavior:
SettingWhat Happens
auto_commit: false (default)Agent creates a new branch and opens a new PR
auto_commit: trueAgent commits directly to the PR branch
triggers:
  - event: pull_request
    auto_commit: false  # Safer: creates separate PR
Use auto_commit: true carefully—it can cause merge conflicts if others are pushing to the same PR.

​
GitHub Check Runs

Every agent execution automatically creates a GitHub Check Run on the triggering commit or PR.

​
Check Run Lifecycle

  1. Queued — Check created when agent starts
  2. In Progress — Agent is executing
  3. Completed — Finished with success or failure

​
What’s Included

  • Title: Agent name (e.g., “documentation_agent”)
  • Summary: Brief description of the task
  • Output: Agent’s detailed output (up to 65,535 characters)
Check runs are visible in:
  • The PR’s “Checks” tab
  • Commit status indicators
  • Branch protection rules (can require checks to pass)
You don’t need to configure anything—check runs are created automatically for every agent execution.

​
Slack Notifications

Send messages to Slack channels using the slack_tool.

​
Enabling Slack

Add slack_tool to your agent configuration: 
tools:
  include:
    - slack_tool
Slack integration must be configured in the Buster platform under Settings → Integrations → Slack.

​
Sending Messages

Reference Slack channels in your prompt: 
prompt: |
  Send a message to #data-alerts:
  "🚨 Data quality issue detected in orders table.
  
  Null rate: 15% (threshold: 5%)
  
  Investigation needed."

​
Message Formatting

Agents can use Slack’s mrkdwn formatting: 
prompt: |
  Send to #data-team:
  
  "*Weekly Documentation Report*
  
  :white_check_mark: 45 models fully documented
  :warning: 12 models need attention
  :x: 3 models have no documentation
  
  <https://example.com/report|View full report>"
Formatting options:
  • *bold* — Bold text
  • _italic_ — Italic text
  • ~strikethrough~ — Strikethrough
  • `code` — Inline code
  • :emoji_name: — Emoji

​
Channel Requirements

The Buster bot must be invited to channels before it can post:
  1. Open the Slack channel
  2. Type /invite @Buster
  3. The bot can now post to that channel

​
Action Patterns

​
Conditional Actions

Take different actions based on findings: 
prompt: |
  Check documentation coverage for changed models.
  
  If coverage < 50%:
    - Post a comment requesting documentation
    - Send alert to #data-team on Slack
  
  If coverage 50-80%:
    - Post a comment with suggestions
  
  If coverage > 80%:
    - Post a brief "looks good" comment

​
Sequential Workflows

Execute actions in a specific order: 
prompt: |
  1. First, analyze all changed models
  2. Generate documentation for each
  3. Run dbt parse to validate
  4. If validation passes:
     - Commit changes
     - Create PR
     - Post summary to Slack
  5. If validation fails:
     - Post error as PR comment
     - Do NOT create PR

​
Error Handling

Define what to do when things go wrong: 
prompt: |
  Update documentation for changed models.
  
  If successful:
    - Commit and push changes
    - Post brief confirmation comment
  
  If dbt parse fails:
    - Post the error message as a PR comment
    - Tag @data-platform for review
    - Do not commit broken YAML
  
  If warehouse is unreachable:
    - Post a comment explaining the issue
    - Skip metadata profiling
    - Create documentation from SQL analysis only

​
Complete Examples

​
Documentation Agent with Notifications

name: docs_with_notifications

prompt: |
  Update documentation for models changed in this PR.
  
  For each changed SQL file:
  1. Use RetrieveMetadata to profile the model
  2. Update or create the YAML documentation
  3. Include:
     - Model description with purpose and grain
     - Column descriptions with data types
     - Null rates for columns with >5% nulls
  
  After processing all models:
  1. Run dbt parse to validate
  2. If valid:
     - Commit with message "docs: Update documentation for [models]"
     - Post PR comment listing what was updated
  3. If invalid:
     - Post the parse error as a PR comment
     - Send alert to #data-platform on Slack
     - Do NOT commit broken YAML

triggers:
  - event: pull_request
    types: ['opened', 'synchronize']
    includes:
      - "models/**/*.sql"
    auto_commit: false

tools:
  include:
    - slack_tool

​
Scheduled Audit with Report

name: weekly_audit

prompt: |
  Perform a weekly documentation audit.
  
  1. Scan all models in models/marts/
  2. Calculate:
     - Overall documentation coverage
     - Models missing descriptions
     - Columns missing descriptions
  
  3. Send report to #data-documentation on Slack:
     "*Weekly Documentation Audit*
     
     :bar_chart: *Coverage*
     Models: X% documented
     Columns: Y% documented
     
     :warning: *Needs Attention*
     [List top 5 undocumented models]
     
     :clipboard: *Action Items*
     - Focus on marts layer this week
     - 3 new models need documentation"
  
  4. If critical models are undocumented:
     - Create a GitHub issue with the list
     - Assign to @data-team

triggers:
  - event: scheduled
    cron: "0 9 * * 1"  # Every Monday at 9 AM UTC
    branches: ['main']

tools:
  include:
    - slack_tool

​
Failure Alert Agent

name: pipeline_alerts

prompt: |
  An Airflow DAG has failed. Investigate and alert the team.
  
  1. Extract error details from the trigger context
  2. Search the repository for related models
  3. Check if this is a recurring issue (search recent issues)
  
  4. Send alert to #data-incidents on Slack:
     ":rotating_light: *Pipeline Failure*
     
     *DAG:* [dag_id]
     *Task:* [task_id]
     *Error:* [error_message]
     
     *Possibly Related:*
     - [List related models or files]
     
     *Next Steps:*
     1. Check warehouse connectivity
     2. Review recent schema changes
     3. Check upstream data freshness"
  
  5. Create a GitHub issue:
     - Title: "Pipeline failure: [dag_id]"
     - Label: "incident"
     - Include error details and investigation links

triggers:
  - event: airflow
    type: dag_run_failed

tools:
  include:
    - slack_tool

​
Best Practices

​
Be Explicit About Actions

Tell the agent exactly what actions to take: 
# Vague
prompt: Let me know if there are issues

# Explicit
prompt: |
  If issues found:
  - Post a PR comment with file:line references
  - Include code examples of the correct pattern
  - Send summary to #data-team on Slack

​
Handle All Outcomes

Cover success, failure, and edge cases: 
prompt: |
  If changes needed:
    - Make changes, commit, create PR
  
  If no changes needed:
    - Post brief "all good" comment
  
  If errors occur:
    - Post error details as comment
    - Send alert to Slack
    - Do NOT commit partial changes

​
Use Consistent Formatting

Define message formats in your prompt: 
prompt: |
  Format PR comments with this structure:
  
  ## [Status Emoji] [Title]
  
  **Summary:** Brief description
  
  **Details:**
  - Bullet points for specifics
  
  **Next Steps:**
  1. Numbered action items

​
Rate Limit Notifications

Avoid spamming channels: 
prompt: |
  Only send Slack notifications for:
  - Critical issues (null rate > 20%)
  - Failed validations
  - Completed batch operations
  
  Do NOT notify for:
  - Routine successful runs
  - Minor warnings