| name | agently-agents-and-prompts |
| description | Agent setup, prompt layering, routing, and info injection patterns. |
Agently Agents and Prompts Skill
Use this skill for agent configuration, prompt layering, YAML-based prompts, and multi-agent routing.
Key Patterns
- Keep stable rules in agent prompts; keep dynamic input in request prompts.
- Use YAML prompts for maintainable configurations.
- Route to specialized agents by intent.
Pitfalls to Avoid (Lessons from NexusTodo)
- Fix action enums and status enums in the system prompt to prevent invalid actions.
- Map common Chinese phrases to structured fields (e.g., "未完成" =>
status_list).
- Rename requests should set
title=NEW and query.keyword=OLD (without "任务" suffix).
- Deletion with keyword should be allowed as bulk without extra clarification.
- For “这些/上述/刚才列出的任务”, return
selection_indices to select by list order.
References
examples/prompt_layers_and_mappings.py
examples/prompt_config_from_yaml.py
examples/multi_agent_router.py
examples/rag_with_info_prompt.py
examples/single_request_multi_read.py
examples/prompt_template.yaml
Examples
See examples/run.sh for runnable commands.