Claude Code subagent imported from savannah-i-g/DryDock (.claude/agents/README.md). Copyright stays with the author.

DryDock Runtime System

Autonomous execution for sophisticated DryDock agents

This directory contains Claude Code subagents that enable autonomous execution of DryDock agent configurations. The runtime system allows you to leverage DryDock's sophisticated agent workflows, decision logic, and quality gates in autonomous, task-completing mode.

What is DryDock Runtime?

DryDock Runtime is an execution engine that:

• Loads DryDock agent configurations from the output/ directory
• Executes their capabilities autonomously without user interaction
• Applies workflows, decision logic, and quality gates
• Returns comprehensive, high-quality results

Think of it as: DryDock agents = sophisticated configuration | Runtime = autonomous executor

Available Subagents

1. drydock-runtime.md

The main autonomous execution engine

• Purpose: Loads and executes DryDock agent configurations autonomously
• Tools: Read, Write, Edit, Bash, Grep, Glob, WebSearch, WebFetch
• Use When: You have a well-defined task and want autonomous execution to completion

Example usage:

"Use drydock-runtime to review this authentication code for security issues"

The runtime will:

1. Identify the task (code review with security focus)
2. Load the appropriate agent (Developer)
3. Extract the code_review capability
4. Apply decision logic (auth code → prioritize security)
5. Execute the workflow autonomously
6. Return comprehensive security review

2. drydock-selector.md

Agent discovery and recommendation helper

• Purpose: Helps you find and select the right DryDock agent
• Tools: Read, Glob
• Use When: You're unsure which agent to use or want to see what's available

Example usage:

"Use drydock-selector to show me what agents I have"
"Use drydock-selector to recommend an agent for project planning"

How It Works

Architecture

User Request
    ↓
Claude Code Subagent (drydock-runtime)
    ↓
Loads DryDock Agent Config (output/[agent]/)
    ↓
Parses: capabilities, workflows, decision logic, quality gates
    ↓
Maps permissions → Claude Code tools
    ↓
Executes workflow autonomously
    ↓
Returns comprehensive result

Execution Flow

1. Task Analysis - Runtime identifies task type and domain
2. Agent Selection - Chooses appropriate DryDock agent
3. Configuration Loading - Reads agent's core configuration file
4. Capability Matching - Finds the right capability for the task
5. Workflow Execution - Follows the workflow pattern (sequential, parallel, iterative, etc.)
6. Quality Enforcement - Applies quality gates at validation points
7. Result Delivery - Returns complete, professional output

Usage Modes

Interactive Mode (Traditional DryDock)

User: Provide output/developer/developer.activationkey
Agent: [Presents menu, asks questions, guides step-by-step]

Best for:

• Exploratory tasks
• When you want control over each step
• Learning agent capabilities
• Building complex configurations interactively

Autonomous Mode (Runtime)

User: "Use drydock-runtime to [task]"
Runtime: [Loads config, executes autonomously, returns result]

Best for:

• Well-defined tasks
• Batch operations
• Autonomous execution to completion
• When you want results without interaction

Both modes use the same agent configuration - you choose based on your needs.

Task-to-Agent Mapping

The runtime automatically selects the right agent based on your task:

Prerequisites

Required: DryDock Agents

The runtime needs DryDock agents to execute. To create agents:

1. Activate DryDock:

   Provide drydock.activationkey to Claude
   

2. Select a template or build custom:

   • Quick Start: Choose from 20 pre-built templates
   • Custom Build: Define your own agent from scratch

3. Generate the agent:

   • DryDock creates files in output/[agent_name]/
   • Includes .activationkey and _core.[extension] files

4. Now use runtime:

   "Use drydock-runtime to [your task]"
   

Agent Status Check

To see what agents you have:

"Use drydock-selector to list available agents"

Examples

Example 1: Code Security Review

User: "Use drydock-runtime to review this authentication function for vulnerabilities"

Runtime Execution:
1. Loads: output/developer/developer_core.developer
2. Capability: code_review
3. Decision Logic: "if_auth_code" → prioritize_security_analysis
4. Workflow: Sequential with security quality gates
5. Tools: Read (code), Grep (patterns), Bash (if needed)
6. Output: Comprehensive security review with findings

Result: Detailed security assessment with:
- Vulnerability analysis (SQL injection, auth bypass, etc.)
- Input validation check
- Session management review
- Specific recommendations

Example 2: Project Planning

User: "Use drydock-runtime to create a project plan for implementing SSO"

Runtime Execution:
1. Loads: output/project_manager/project_manager_core.projectmanager
2. Capability: project_planning
3. Workflow: Coordinator with task decomposition
4. Tools: Write (plan document), Read (context)
5. Quality Gates: Completeness, realistic estimates, clear dependencies
6. Output: Complete project plan

Result: Detailed plan with:
- Timeline and milestones
- Task breakdown with dependencies
- Resource allocation
- Risk assessment
- Success metrics

Example 3: API Documentation

User: "Use drydock-runtime to generate API docs for the user service"

Runtime Execution:
1. Loads: output/technical_writer/technical_writer_core.technicalwriter
2. Capability: api_documentation
3. Workflow: Sequential with clarity and completeness gates
4. Tools: Read (analyze code), Write (generate docs)
5. Quality Gates: Clear descriptions, complete examples, proper formatting
6. Output: Complete API documentation

Result: Professional docs with:
- Endpoint descriptions
- Request/response schemas
- Authentication details
- Code examples
- Error handling

Integration with DryDock

Preservation of Existing System

The runtime is 100% additive - no changes to existing DryDock:

• All 20 templates unchanged
• Core builder logic unchanged
• Interactive Q&A workflow unchanged
• Validation engine unchanged
• File generation unchanged

New Addition:

• library/tool_mappings.library - Maps permissions to tools (reference only)
• .claude/agents/ - Runtime subagents (optional feature)

Coexistence

Both modes work with the same agent configurations:

Same Agent, Two Modes:

• Interactive: Activate .activationkey → Menu-driven Q&A
• Autonomous: Invoke runtime → Autonomous execution

Choice Based on Task:

• Exploratory, learning → Interactive mode
• Well-defined, batch → Autonomous mode

Technical Details

Permission-to-Tool Mapping

DryDock agents define abstract permissions. Runtime maps them to Claude Code tools:

See library/tool_mappings.library for complete mappings

Workflow Execution

Runtime handles all DryDock workflow patterns:

1. Sequential - Steps in order
2. Parallel - Concurrent execution
3. Iterative - Refinement loops
4. Conditional - Branch-based routing
5. Pipeline - Multi-stage with gates
6. Coordinator - Sub-workflow orchestration
7. Feedback Loop - Continuous improvement
8. Event-Driven - Condition-based response

Quality Gate Enforcement

Runtime applies quality gates from agent configs:

Security Gates:

• No SQL injection
• Proper authentication
• Input validation

Performance Gates:

• No N+1 queries
• Efficient algorithms
• Optimized resources

Maintainability Gates:

• Clear naming
• Adequate documentation
• Modular design

Troubleshooting

Agent Not Found

WARNING: The [Agent Name] agent hasn't been built yet.

Solution:
1. Activate DryDock: provide drydock.activationkey
2. Select the appropriate template
3. Generate the agent
4. Retry your task

Unexpected Behavior

• Check agent config is valid (well-formed JSON)
• Ensure all required tools are available
• Verify agent has the needed capability
• Review runtime execution logs

Performance

• Runtime caches loaded configs per session
• Reuses agent config for multiple tasks
• Reloads only when different agent needed

Future Enhancements

Planned Features:

• Multi-agent orchestration (coordinate multiple DryDock agents)
• Learning system (improve based on execution history)
• Visual workflow builder (GUI for agent workflows)
• Performance analytics (track agent execution metrics)
• Community marketplace (share runtime-compatible agents)

Contributing

To extend the runtime system:

1. Add new tool mappings:

   • Edit library/tool_mappings.library
   • Add permission → tool mappings
   • Document usage patterns

2. Enhance runtime logic:

   • Modify .claude/agents/drydock-runtime.md
   • Add new workflow patterns
   • Improve decision logic

3. Create helper subagents:

   • Add specialized helpers to .claude/agents/
   • Follow same YAML frontmatter format
   • Document purpose and usage

License

DryDock Runtime is part of the DryDock project.

GPL-3.0 Licensed - Free to use, modify, and distribute with attribution.

Copyright 2025 Savannah Goring

Quick Reference

List available agents:

"Use drydock-selector to show available agents"

Get agent recommendation:

"Use drydock-selector to recommend an agent for [task]"

Execute task autonomously:

"Use drydock-runtime to [your task]"

Interactive mode:

Provide: output/[agent]/[agent].activationkey