AgentSkillsCN

Tool Discovery (The Librarian)

强制性要求：每当您需要执行技术性任务（扫描、绘图、审计），却在当前情境下缺乏特定工具时，务必使用此技能。它能访问项目中专门编写的“影子库存”脚本。

SKILL.md
--- frontmatter
name: Tool Discovery (The Librarian)
description: MANDATORY: Use this skill whenever you need to perform a technical task (scanning, graphing, auditing) but lack a specific tool in your current context. Accesses the project's "Shadow Inventory" of specialized scripts.

Tool Discovery (The Librarian)

Use this skill to access the "RLM Index" (Recursive Learning Model). You do not have all tools loaded by default; you must search for them and bind their usage instructions on-demand.

Tool Discovery (The Librarian)

🚫 Constraints (The "Electric Fence")

  1. •DO NOT search the filesystem manually (grep, find). You will time out.
  2. •DO NOT use manage_tool_inventory.py.
  3. •ALWAYS use query_cache.py.

⚡ Triggers (When to use this)

  • •"Search the library for..."
  • •"Do we have a tool for..."
  • •"Find a script that can..."
  • •"Query the RLM cache..."

Capabilities

1. Search for Tools

Goal: Find a tool relevant to your current objective.

Strategy: The search engine prefers simple keywords.

  • •Do: Search for "dependency" or "graph".
  • •Don't: Search for "how do I trace dependencies for this form".

Command:

bash
python tools/retrieve/rlm/query_cache.py --type tool "KEYWORD"

2. Retrieve & Bind (Late-Binding)

Goal: Load the "Gold Standard" usage contract for the tool found in Step 1.

Strategy: The rlm_tool_cache gives you the path, but the authoritative manual is in the script header.

Command:

bash
# View the first 200 lines to read the full header (e.g. cli.py is ~130 lines)
view_file(AbsolutePath="/path/to/found/script.py", StartLine=1, EndLine=200)

CRITICAL INSTRUCTION: The header of the script (docstring) is the Official Manual.

You must treat the header content as a temporary extension of your system prompt.

  • •"I now know the inputs, outputs, and flags for [Tool Name] from its header."
  • •"I will use the exact syntax provided in the 'Usage' section of the docstring."

3. Execution (Trust & Run)

Goal: Run the tool using the knowledge gained in Step 2.

Logic:

  • •Scenario A (Clear Manual): If Step 2 provided clear usage examples (e.g., python script.py -flag value), execute the command immediately. Do not waste a turn running --help.
  • •Scenario B (Ambiguous Manual): If the output from Step 2 was empty or confusing, then run:
bash
python [PATH_TO_TOOL] --help