discover-agentic（发现 Agent）

Agentic Workflow Skills Discovery

Provides automatic access to skills for building autonomous AI agents, managing tool use, and orchestrating multi-step workflows.

When This Skill Activates

This skill auto-activates when you're working with:

• AI agent development and autonomous workflows
• Task decomposition for LLM agents
• Function calling and tool use patterns
• Context window management and summarization
• Working memory, scratchpads, and state tracking
• Long-term memory and persistence patterns
• Multi-step reasoning and planning
• Failure recovery and re-planning strategies
• Parallel vs sequential execution decisions

Available Skills

Quick Reference

The Agentic category contains 3 specialized skills:

1. agentic-task-decomposition - Breaking complex goals into agent-executable steps, dependency graphs, re-planning
2. agentic-tool-use - Function calling patterns, parallel tool use, error handling, tool selection heuristics
3. agentic-memory - Context window management, working memory, long-term persistence, summarization

Load Full Category Details

For complete descriptions and workflows:

Read../agentic/INDEX.md

This loads the full Agentic category index with:

• Detailed skill descriptions
• Usage triggers for each skill
• Common workflow combinations
• Cross-references to related skills

Load Specific Skills

Load individual skills as needed:

Task planning and decomposition

Read../agentic/agentic-task-decomposition.md

Tool calling and execution

Read../agentic/agentic-tool-use.md

Context and memory management

Read../agentic/agentic-memory.md

Common Workflows

Building an Agent from Scratch

Sequence: Decomposition -> Tool Use -> Memory

Read../agentic/agentic-task-decomposition.md # Plan the agent's task handling Read../agentic/agentic-tool-use.md # Implement tool execution Read../agentic/agentic-memory.md # Add context management

Adding Tool Use to an Agent

Sequence: Tool Use -> Decomposition

Read../agentic/agentic-tool-use.md # Tool calling patterns Read../agentic/agentic-task-decomposition.md # Multi-tool task planning

Scaling Agent Sessions

Sequence: Memory -> Decomposition

Read../agentic/agentic-memory.md # Context management Read../agentic/agentic-task-decomposition.md # Context-bounded task sizing

Complete Agentic Stack

Full implementation:

1. Planning layer

Read../agentic/agentic-task-decomposition.md

2. Execution layer

Read../agentic/agentic-tool-use.md

3. Memory layer

Read../agentic/agentic-memory.md

Skill Selection Guide

Choose task decomposition when:

• Complex goal needs breaking into steps
• Tasks have dependencies or ordering constraints
• Need to handle ambiguous requirements
• Building a planning layer for an agent

Choose tool use when:

• Implementing function calling in an LLM agent
• Debugging tool call failures or inefficiency
• Designing tool schemas and structured outputs
• Optimizing multi-tool workflows

Choose memory when:

• Agent sessions are hitting context limits
• Need to persist state across turns or sessions
• Building multi-turn conversational agents
• Implementing retrieval-augmented generation

Integration with Other Skills

Agentic skills commonly combine with:

API skills (discover-api):

• Agents that call REST/GraphQL APIs
• Tool schemas for API endpoints
• Authentication handling in agent workflows

Testing skills (discover-testing):

• Testing agent behaviors and tool use
• Autonomous test generation
• Verifying agent task completion

Database skills (discover-database):

• Database-backed agent memory
• Agents that query and modify data
• Persistent state management

Infrastructure skills (discover-infra, discover-cloud):

• Deploying agent systems
• Scaling agent workloads
• Cost optimization for LLM calls

Usage Instructions

1. Auto-activation: This skill loads automatically when Claude Code detects agentic workflow tasks
2. Browse skills: Run Read../agentic/INDEX.md for full category overview
3. Load specific skills: Use commands above to load individual skills
4. Follow workflows: Use recommended sequences for common agentic patterns
5. Combine skills: Load multiple skills for comprehensive coverage

Progressive Loading

This gateway skill (~200 lines, ~2K tokens) enables progressive loading:

• Level 1: Gateway loads automatically (you're here now)
• Level 2: Load category INDEX.md (~3K tokens) for full overview
• Level 3: Load specific skills (~2-3K tokens each) as needed

Total context: 2K + 3K + skill(s) = 5-10K tokens vs 12K+ for entire category.

Quick Start Examples

"Build an autonomous coding agent": Read../agentic/agentic-task-decomposition.md

"How should my agent call tools?": Read../agentic/agentic-tool-use.md

"My agent is running out of context": Read../agentic/agentic-memory.md

Next Steps: Run Read../agentic/INDEX.md to see full category details, or load specific skills using the commands above.