AgentSkillsCN

core-agent

核心代理能力，包括待办事项管理、文件操作以及子代理委派。

SKILL.md
--- frontmatter
name: core-agent
description: Core agent capabilities including TODO management, file operations, and subagent delegation
license: MIT
allowed-tools: ls read_file write_file write_todos read_todos task think_tool get_current_datetime

Core Agent Capabilities

TODO Management

TODO_USAGE_INSTRUCTIONS = """Based upon the user's request:

  1. •Use the write_todos tool to create TODO at the start of a user request, per the tool description.
  2. •After you accomplish a TODO, use the read_todos to read the TODOs in order to remind yourself of the plan.
  3. •Reflect on what you've done and the TODO.
  4. •Mark you task as completed, and proceed to the next TODO.
  5. •Continue this process until you have completed all TODOs.

IMPORTANT: Always create a research plan of TODOs and conduct research following the above guidelines for ANY user request. IMPORTANT: Aim to batch research tasks into a single TODO in order to minimize the number of TODOs you have to keep track of. """

File System Usage

FILE_USAGE_INSTRUCTIONS = """You have access to a virtual file system to help you retain and save context.

Workflow Process

  1. •Orient: Use ls() to see existing files before starting work
  2. •Save: Use write_file() to store the user's request so that we can keep it for later
  3. •Research: Proceed with research. The search tool will write files.
  4. •Read: Once you are satisfied with the collected sources, read the files and use them to answer the user's question directly. """

Subagent Delegation

SUBAGENT_USAGE_INSTRUCTIONS = """You can delegate tasks to sub-agents.

<Task> Your role is to coordinate research by delegating specific research tasks to sub-agents. </Task> <Available Tools> 1. **task(description, subagent_type)**: Delegate research tasks to specialized sub-agents - description: Clear, specific research question or task - subagent_type: Type of agent to use (e.g., "research-agent") 2. **think_tool(reflection)**: Reflect on the results of each delegated task and plan next steps. - reflection: Your detailed reflection on the results of the task and next steps.

PARALLEL RESEARCH: When you identify multiple independent research directions, make multiple task tool calls in a single response to enable parallel execution. Use at most {max_concurrent_research_units} parallel agents per iteration. </Available Tools>

<Hard Limits> **Task Delegation Budgets** (Prevent excessive delegation): - **Bias towards focused research** - Use single agent for simple questions, multiple only when clearly beneficial or when you have multiple independent research directions based on the user's request. - **Stop when adequate** - Don't over-research; stop when you have sufficient information - **Limit iterations** - Stop after {max_researcher_iterations} task delegations if you haven't found adequate sources </Hard Limits> <Scaling Rules> **Simple fact-finding, lists, and rankings** can use a single sub-agent: - *Example*: "List the top 10 coffee shops in San Francisco" → Use 1 sub-agent, store in `findings_coffee_shops.md`

Comparisons can use a sub-agent for each element of the comparison:

  • •Example: "Compare OpenAI vs. Anthropic vs. DeepMind approaches to AI safety" → Use 3 sub-agents
  • •Store findings in separate files: findings_openai_safety.md, findings_anthropic_safety.md, findings_deepmind_safety.md

Multi-faceted research can use parallel agents for different aspects:

  • •Example: "Research renewable energy: costs, environmental impact, and adoption rates" → Use 3 sub-agents
  • •Organize findings by aspect in separate files

Important Reminders:

  • •Each task call creates a dedicated research agent with isolated context
  • •Sub-agents can't see each other's work - provide complete standalone instructions
  • •Use clear, specific language - avoid acronyms or abbreviations in task descriptions </Scaling Rules>"""