AgentSkillsCN

hekate-agent-workflow

Hekate 多智能体工作流的模式与规范。

SKILL.md
--- frontmatter
name: hekate-agent-workflow
description: Hekate multi-agent workflow patterns and conventions

Hekate Agent Workflow

Task Completion Convention

When completing a task, commit with explicit marker:

bash
git add .
git commit -m "Complete task-abc123: Implemented feature X"

This triggers posttooluse_complete_task.py to mark task as complete in Redis.

Epic Creation

bash
create epic: Build authentication system

Triggers userpromptsubmit_decompose.py to:

  1. Call OpenRouter API for task decomposition
  2. Create tasks in Beads with complexity ratings
  3. Store epic metadata in Redis
  4. Assign providers based on complexity

Provider Assignment

ComplexityProviderPool Size
1-4 (low)DeepSeek6 agents
5-7 (medium)GLM4 agents
8-10 (high)Claude2 agents

Subagent Execution

When assigned a task:

  1. Check Redis for task details: redis-cli GET "task:{id}:description"
  2. Execute implementation
  3. Commit with "Complete task-{id}" marker
  4. Hook auto-detects completion

Memory System

Store Solutions

After solving a problem, the solution is automatically stored in semantic memory via posttooluse_memory.py.

Retrieve Related Solutions

pretooluse_memory.py injects similar past solutions when starting new tasks.

Verification

High-complexity tasks trigger posttooluse_verify_prefetch.py to start verification in parallel.

Troubleshooting

Task stuck in pending

bash
bd show {task-id}
redis-cli GET "task:{task-id}:status"

Agent not spawning

Check posttooluse_spawn_agents.py logs in ~/.hekate/logs/

Quota exhausted

bash
redis-cli SET "quota:claude:count" "0"