Agents

Maestro uses specialized agents for each stage of the pipeline. Each agent has its own system prompt, model configuration, and execution logic.

Implementation Agent

Purpose: Write code to fulfill the task requirements.

How it works:

1. Reads the task description and any linked issues
2. Explores the project codebase to understand structure, conventions, and patterns
3. Writes the implementation — new files, modified files, tests
4. Creates a Git branch and opens a pull request
5. Attaches code snippets and file paths to the task activity log

Configuration:

• model — which LLM to use (default: Claude)
• system_prompt — custom instructions for the agent
• max_file_reads — limit on how many files the agent can read per run
• branch_prefix — prefix for created branches (default: maestro/)

Review Agent

Purpose: Perform inline code review on pull requests.

How it works:

1. Fetches the PR diff from GitHub
2. Reads each changed file in context
3. Leaves inline comments on specific lines where issues are found
4. Comments are categorized: bug, style, performance, security, suggestion
5. If no issues found, approves the review

Configuration:

• model — which LLM to use
• system_prompt — review criteria and style guidelines
• severity_threshold — minimum severity to flag (default: suggestion)
• max_comments — cap on number of comments per review

Risk Profile Agent

Purpose: Assess the risk of merging a change.

How it works:

1. Analyzes the PR diff for complexity metrics
2. Checks which modules and dependencies are affected (blast radius)
3. Verifies test coverage for changed code
4. Produces a risk score: LOW, MEDIUM, or HIGH
5. LOW risk tasks proceed automatically; MEDIUM and HIGH require human approval

Configuration:

• model — which LLM to use
• system_prompt — risk assessment criteria
• auto_merge_threshold — risk level that can auto-merge (default: LOW)

Deployment Agent

Purpose: Merge pull requests and monitor CI.

How it works:

1. Merges the approved PR into the target branch
2. Monitors CI pipeline status (GitHub Actions, etc.)
3. Reports success or failure
4. On failure, attaches CI logs to the task

Configuration:

• merge_strategy — squash, merge, or rebase (default: squash)
• require_ci_pass — wait for CI before marking as deployed (default: true)
• target_branch — branch to merge into (default: main)

Monitor Agent

Purpose: Post-deployment health monitoring.

How it works:

1. Watches application logs and error rates after deployment
2. Compares error rates to pre-deployment baseline
3. Flags anomalies if error rates spike
4. Can trigger rollback alerts

Configuration:

• monitoring_window — how long to watch after deploy (default: 30m)
• error_threshold — percentage increase that triggers an alert (default: 50%)
• log_sources — which log streams to watch

Custom agents

You can create custom agents using the plugin framework. See Plugins for details.