Agent Guide

Mission Clawtrol ships with 9 specialized AI agents, each designed for a specific role in the software development lifecycle. This guide explains what each agent does, how to assign tasks to them, and how delegation works.

The 9-Agent Roster

🎯 CSO — Chief Strategy Officer

Role: Orchestration, planning, and delegation.

The CSO is your top-level agent. When you assign a task to the CSO, it:

Reads the task description and project context
Determines which specialist agent(s) should handle it
Breaks it into sub-tasks if needed
Delegates to the appropriate agents
Reviews the results and moves the task forward

The CSO is the right choice when you’re not sure which agent to use, or when a task spans multiple roles.

Best for: High-level feature requests, multi-step tasks, anything that needs coordination.

💻 Senior Developer

Role: Architecture, complex features, code review.

The Senior Dev handles technically challenging work: designing systems, writing complex business logic, reviewing code for quality and correctness, and mentoring Junior Dev output.

Best for: New features, system design, architectural decisions, pull request reviews, performance optimization.

Capabilities:

Reads and writes code across any language
Commits to git with meaningful commit messages
Runs tests and interprets results
Reviews Junior Dev work before marking tasks complete

🔧 Junior Developer

Role: Implementation, bug fixes, tests.

The Junior Dev handles well-defined, straightforward implementation tasks. It’s faster and cheaper than the Senior Dev for routine work.

Best for: Bug fixes, simple features with clear specs, writing unit tests, code refactoring with clear instructions.

Capabilities:

Reads and writes code
Commits to git
Runs tests
Escalates to Senior Dev if the task exceeds its scope

🔬 Senior Researcher

Role: Deep research, analysis, competitive intelligence.

The Senior Researcher performs thorough, structured research. It reads extensively, synthesizes information, and produces well-cited, analytical output.

Best for: Competitive analysis, technology evaluation, literature review, market research, architecture decision records (ADRs).

Capabilities:

Web search and page reading
Structured report writing
Source citation
Long-form analysis (10+ pages when needed)

📚 Junior Researcher

Role: Data gathering, fact-checking, quick lookups.

The Junior Researcher handles faster, narrower research tasks. It’s cost-effective for gathering data, checking facts, or pulling together a summary.

Best for: Quick fact checks, data gathering for a specific question, summarizing a single document or article.

✍️ Editor

Role: Documentation, blog posts, copy, communications.

The Editor writes and refines written content. It’s not a developer — don’t give it code tasks. It excels at making technical content readable, polished, and well-structured.

Best for: README files, blog posts, API documentation (prose), changelog entries, Slack announcements, marketing copy.

Capabilities:

Writes and edits Markdown
Follows style guides (provide one in the task or PROJECT.md)
Adapts tone (technical, casual, formal)

🧪 QA Agent

Role: Test plans, quality reviews, done-criteria checks.

The QA agent has two primary functions:

Test planning — given a feature description, write a comprehensive test plan
Done-criteria review — automatically runs when a task moves to the Review column, checking whether the work meets the stated acceptance criteria

Best for: Writing test plans, reviewing completed tasks, writing QA documentation.

Automated QA: When any task moves to “Review” status, the QA agent automatically reviews it against the task’s done criteria. If it passes, the task closes. If not, it’s flagged with feedback.

🛡️ Security Auditor

Role: Security audits, vulnerability reviews.

The Security Auditor reviews code, configurations, and architectures for security issues. It checks against OWASP guidelines, identifies common vulnerabilities, and recommends mitigations.

Best for: Pre-deployment security reviews, reviewing auth implementations, checking for injection vulnerabilities, dependency audits.

📋 Product Manager

Role: Scope, roadmap, stakeholder reports.

The PM agent handles the business and planning side of your project. It writes specifications, breaks down epics into tasks, maintains roadmaps, and generates reports.

Best for: Breaking down feature requests into tasks, writing PRDs, generating the weekly stakeholder report, maintaining the project roadmap.

Assigning Tasks

Direct Assignment

Select a specific agent when you know who should handle the work:

Click New Task
Set Assign to: Senior Dev (or any agent)
Click Assign & Run

The agent starts immediately.

CSO Delegation

Assign to the CSO when the task is complex or cross-cutting:

Click New Task
Set Assign to: CSO
Click Assign & Run

The CSO reads the task, decides how to handle it, and delegates to one or more specialists. You’ll see sub-tasks created and assigned automatically.

Writing Good Task Descriptions

Agents perform much better with clear, specific task descriptions. Include:

What to do — the specific deliverable
Context — relevant background, linked files, or previous decisions
Done criteria — how to know when it’s finished
Constraints — anything they should NOT do

Good example:

Title: Add password reset flow

Description:
Implement the full password reset flow for the auth system.

Technical details:
- Use the existing User model (src/models/user.ts)
- Send reset emails via the existing email service (src/services/email.ts)
- Token should expire after 1 hour
- Store hashed reset tokens, never plaintext

Done when:
- /auth/forgot-password endpoint accepts email, sends reset email
- /auth/reset-password endpoint validates token, updates password
- Old reset tokens invalidated after use
- Unit tests cover happy path + expired token + invalid token

How Delegation Works

When the CSO receives a task, it follows this process:

CSO receives task
     ↓
Reads PROJECT.md + task description
     ↓
Determines complexity and scope
     ↓
Single agent? → Direct assignment
Multi-agent? → Creates sub-tasks, assigns each
     ↓
Monitors progress
     ↓
Reviews output
     ↓
Marks parent task complete or escalates

Sub-tasks appear on the board under the parent task, each with its own assigned agent and status.

Agent Configuration

Each agent’s behavior can be customized from the Agents page:

Setting	Description
Model	Which AI model powers this agent (claude-haiku, claude-sonnet, claude-opus)
System prompt	Custom instructions appended to the agent’s default persona
Max tokens	Limit response length (useful for cost control)
Enabled	Toggle an agent on/off

Agent configs are stored in agents/config.json and can be committed to version control.