CCPM Agents

CCPM Agents

Specialized subagents for CCPM operations and development workflows.

Overview

CCPM agents are specialized components that handle specific domains of functionality. They are invoked automatically by commands, hooks, or explicitly via the Task tool, providing focused expertise and optimized token usage through caching and batching.

Agent Categories

Linear and PM Operations

These agents handle project management integrations and provide centralized API operations with caching.

Agent	Purpose
linear-operations	Central handler for all Linear API operations with session-level caching
pm-operations-orchestrator	Lightweight coordinator for multi-PM system operations with lazy loading
jira-operations	Jira API operations with caching and Markdown/ADF conversion
confluence-operations	Confluence API operations with content-aware caching

Project Management

These agents manage project detection, configuration, and context across CCPM workflows.

Agent	Purpose
project-detector	Detects active project and subproject from git remote, working directory, or patterns
project-config-loader	Loads and validates project configuration from CCPM config file
project-context-manager	Manages active project context including setting, displaying, and switching projects

Development Agents

These agents handle implementation tasks across different technology domains.

Agent	Purpose
frontend-developer	React/UI implementation with design system integration and Tailwind styling
backend-architect	API design, NestJS implementation, database operations, and authentication
tdd-orchestrator	Test-driven development workflow orchestration with coverage requirements
code-reviewer	Automated code review covering security, bugs, performance, and quality
code-quality-enforcer	Automated quality validation running lint/type/test checks on changed files
debugger	Systematic debugging and issue investigation with root cause analysis
security-auditor	Security vulnerability assessment covering OWASP Top 10 and compliance

Design Agents

Agent	Purpose
pm:ui-designer	UI/UX design with wireframes, design system analysis, and developer specifications

CCPM Support Agents

These agents help users work with Claude Code and extend CCPM functionality.

Agent	Purpose
claude-code-guide	Answers questions about Claude Code features, settings, hooks, skills, and MCP
ccpm-developer	Creates and extends CCPM commands, agents, skills, and hooks
ccpm-troubleshooter	Diagnoses and resolves Claude Code and CCPM configuration issues

Agent Selection

Automatic Selection

The smart-agent-selector hook automatically scores and suggests agents based on:

• Keyword matching: +10 per matching keyword in user message
• Task type alignment: +20 for task type relevance
• Tech stack relevance: +15 for matching technologies
• Plugin agent bonus: +5 for CCPM agents
• Project-specific bonus: +25 for project-configured agents

Agents scoring above threshold are automatically invoked or suggested.

Manual Invocation

Invoke agents explicitly using the Task tool:

Task({
  subagent_type: 'ccpm:frontend-developer',
  prompt: `
## Task
Create login form component

## Context
- Issue: PSN-123
- Branch: feature/psn-123-auth

## Requirements
- TypeScript strict mode
- Tailwind styling
- Accessible (a11y)
`
});

Command Delegation

Commands delegate to agents internally based on task type:

Task Type	Agent
UI/Frontend	frontend-developer
API/Backend	backend-architect
Testing	tdd-orchestrator
Code Review	code-reviewer
Quality Checks	code-quality-enforcer
Debugging	debugger
Security	security-auditor
Design	pm:ui-designer

Linear Operations

The linear-operations agent is the central handler for all Linear API interactions. It provides significant token reduction through caching and batching.

Key Benefits

• 50-60% token reduction compared to direct MCP calls
• 85-95% cache hit rate for teams, labels, statuses
• Automatic parameter transformation (handles issueId vs id)
• Structured error handling with suggestions

Critical: Parameter Names

The Linear MCP uses specific parameter names that differ between operations:

Tool	Parameter	Example
get_issue	id	{ id: "WORK-26" }
update_issue	id	{ id: "WORK-26", state: "..." }
create_comment	issueId	{ issueId: "WORK-26", body: "..." }
list_comments	issueId	{ issueId: "WORK-26" }
create_issue	team, title	{ team: "Engineering", title: "..." }

Background Operations

For non-blocking updates, use background execution:

# Post comment (fire-and-forget)
./scripts/linear-background-ops.sh comment PSN-123 "Progress update"

# Update status (fire-and-forget)
./scripts/linear-background-ops.sh update-status PSN-123 "In Progress"

Use blocking calls only when you need the result:

• get_issue - Need the data to continue
• create_issue - Need the issue ID
• update_checklist_items - Need progress for display

PM Operations Orchestrator

The pm-operations-orchestrator provides multi-PM system coordination with:

• Lazy loading: 89% token reduction vs eager loading
• Parallel execution: 50%+ speedup for independent operations
• Dependency management: Correct ordering of dependent operations
• Unified caching: Coordinated cache across all PM systems

Operations

Operation	Purpose
lazy_gather_context	Gather context from multiple PM systems
smart_delegate	Route single operation to appropriate subagent
batch_parallel_execute	Execute multiple operations with dependencies
cache_status	Report unified cache metrics

Creating Agents

Agent Structure

# Agent Name

**Specialized agent for [specific domain]**

## Purpose

What this agent does and why.

## Capabilities

- Capability 1
- Capability 2

## Input Contract

```yaml
task:
  type: string
  description: string
context:
  issueId: string?

Output Contract

result:
  status: "success" | "partial" | "blocked"
  filesModified: string[]
  summary: string

Implementation Patterns

Code examples and patterns.

Integration with CCPM

How commands invoke this agent.

Examples

Concrete usage examples.

### Registration

1. Create agent file in `agents/` directory
2. Add to `.claude-plugin/plugin.json` agents array:
   ```json
   {
     "agents": [
       "./agents/existing-agent.md",
       "./agents/new-agent.md"
     ]
   }

3. Optionally add to smart-agent-selector for auto-invocation

Best Practices

For Agent Design

• Single responsibility: Each agent handles one domain
• Clear contracts: Define explicit input/output YAML schemas
• Implementation patterns: Include code examples agents can follow
• Error handling: Return structured errors with suggestions
• Examples: Provide concrete usage examples

For Agent Usage

• Use Linear subagent: Route all Linear operations through linear-operations
• Delegate to specialists: Use domain-specific agents rather than doing everything in main context
• Provide context: Include issue ID, branch, and relevant technical context
• Handle errors: Check result status and handle failures gracefully

For Performance

• Cache operations: Leverage agent caching for repeated lookups
• Background execution: Use fire-and-forget for non-critical operations
• Parallel tasks: Use parallel Task calls for independent operations
• Minimal context: Keep prompts focused to reduce token usage

Agent Communication

Commands
    │
    ├─> linear-operations (Linear API)
    │
    ├─> pm-operations-orchestrator
    │       ├─> linear-operations
    │       ├─> jira-operations
    │       └─> confluence-operations
    │
    ├─> Development Agents
    │       ├─> frontend-developer
    │       ├─> backend-architect
    │       ├─> tdd-orchestrator
    │       └─> ...
    │
    └─> Project Agents
            ├─> project-detector
            ├─> project-config-loader
            └─> project-context-manager

Summary

Category	Agent Count	Purpose
Linear/PM Operations	4	API operations with caching
Project Management	3	Project detection and context
Development	7	Implementation across domains
Design	1	UI/UX design assistance
CCPM Support	3	Claude Code and CCPM help
Total	18