agentskills.codes
LO

local-tool-reliability

by johnml1135ProductivitySource

A guide to diagnosing and stabilizing local-model tool execution by adjusting gateway settings, prompt size, and tool definitions.

DebuggingAgentsNode.js

Install

mkdir -p .claude/skills/local-tool-reliability && curl -L -o skill.zip "https://agentskills.codes/api/skills/download/17931" && unzip -o skill.zip -d .claude/skills/local-tool-reliability && rm skill.zip

Installs to .claude/skills/local-tool-reliability

Activation

This is the description your AI agent reads to decide when to run this skill — the better it matches your request, the more reliably it fires.

Use when local Copilot/Qwen tool calls fail, malformed tool JSON appears, tools are skipped, or LLM Gateway agent mode is flaky.
128 chars✓ has a “when” trigger
Intermediate

Key capabilities

  • →Separate server health from agent-tool health
  • →Check local-model gateway settings
  • →Check Qwen llama-server launch settings
  • →Trim the tool surface to reduce complexity
  • →Reduce the prompt to isolate issues

How it works

This skill provides a diagnostic procedure to debug and stabilize local-model tool calling by systematically checking server health, gateway settings, Qwen launch configurations, and tool surface complexity.

Inputs & outputs

You give it
A tool-call symptom or error message from a local LLM setup
You get back
An identification of the failure's source (server, gateway, parser, tool-set, or context-size), a suggested reliability setting change, and a validation step

When to use local-tool-reliability

  • →Troubleshoot tool-calling failures in local LLM setups
  • →Diagnose malformed tool JSON output from local models
  • →Optimize tool definitions to improve model reliability
  • →Resolve connection issues between gateway agents and local servers

About this skill

Local Tool Reliability

Debug and stabilize local-model tool calling before changing application code.

Procedure

  1. Separate server health from agent-tool health.
    • Confirm llama-server is running and /v1/models responds.
    • Send a tiny direct chat-completions request to confirm basic generation.
    • If the server is healthy but Copilot fails, suspect gateway, prompt size, or tool-set bloat next.
  2. Check local-model gateway settings.
    • github.copilot.llm-gateway.agentTemperature: 0.0 for tool mode.
    • Start with parallelToolCalling: false for Qwen.
    • Keep enableToolCalling: true unless isolating a basic chat failure.
    • Use a long enough requestTimeout for large local prompts.
  3. Check Qwen llama-server launch settings.
    • Use Jinja chat templates (--jinja).
    • Preserve Qwen thinking when needed for multi-turn tool use (preserve_thinking: true).
    • Use the qwen3_coder tool parser where available.
    • Keep KV cache at q8_0 when reliability matters.
  4. Trim the tool surface.
    • Disable MCP servers and extension tools the task does not need.
    • Prefer one well-named tool over overlapping tools.
    • Make argument names and types unambiguous.
    • Use absolute paths.
    • Prefer plain-text or search-replace edit formats over giant JSON payloads.
  5. Reduce the prompt before blaming the parser.
    • Start a fresh chat and retry with a minimal reproduction.

Output Contract

Return:

  • Whether the failure is server, gateway, parser, tool-set, or context-size.
  • The smallest reliability setting change to try.
  • A validation step.
  • Whether the change trades speed for correctness.

References

When not to use it

  • →When the issue is with application code logic
  • →When basic chat generation is failing due to server unresponsiveness
  • →When the problem is not related to local-model tool calling

Limitations

  • →Focuses only on local-model tool calling issues
  • →Does not address issues with application code
  • →Requires manual execution of diagnostic steps

How it compares

This skill offers a structured troubleshooting guide specifically for local LLM tool-calling issues, which is more targeted than general debugging approaches.

Compared to similar skills

local-tool-reliability side by side with the closest alternatives in the catalog.

SkillInstallsUpdatedSafetyDifficulty
local-tool-reliability (this skill)02moNo flagsIntermediate
langsmith-fetch66moReviewIntermediate
clawhub252moReviewIntermediate
agent-swarm35moNo flagsAdvanced

Try saying

Example prompts that trigger this skill in your AI assistant.

You might also like

langsmith-fetch

ComposioHQ

Debug LangChain and LangGraph agents by fetching execution traces from LangSmith Studio. Use when debugging agent behavior, investigating errors, analyzing tool calls, checking memory operations, or examining agent performance. Automatically fetches recent traces and analyzes execution patterns. Requires langsmith-fetch CLI installed.

67

clawhub

openclaw

Use the ClawHub CLI to search, install, update, and publish agent skills from clawhub.com. Use when you need to fetch new skills on the fly, sync installed skills to latest or a specific version, or publish new/updated skill folders with the npm-installed clawhub CLI.

25151

agent-swarm

ruvnet

Agent skill for swarm - invoke with $agent-swarm

322

reasoningbank-intelligence

ruvnet

Implement adaptive learning with ReasoningBank for pattern recognition, strategy optimization, and continuous improvement. Use when building self-learning agents, optimizing workflows, or implementing meta-cognitive systems.

11

clui-cc-claude-overlay

Aradotso

Command Line User Interface for Claude Code — a floating macOS desktop overlay with multi-tab sessions, permission approval UI, voice input, and skills marketplace.

00

gemini

annex-ai

Gemini CLI for one-shot Q&A, summaries, and generation.

00

Search skills

Search the agent skills registry