Agent Skill
2/7/2026

deep-performance-investigation

[UTILITY] Targeted profiling to evaluate impact of uncertain findings. Called by broad-performance-investigation when confidence is LOW. Returns: expected impact estimate OR clear falsification. NOT a standalone phase.

A
antonykamp
0GitHub Stars
1Views
npx skills add antonykamp/cc-truffle-performance-plugin

SKILL.md

Namedeep-performance-investigation
Description[UTILITY] Targeted profiling to evaluate impact of uncertain findings. Called by broad-performance-investigation when confidence is LOW. Returns: expected impact estimate OR clear falsification. NOT a standalone phase.

Truffle Performance Plugin

A Claude Code plugin for analyzing and optimizing GraalVM Truffle performance. This plugin provides specialized skills for profiling, tracing, and diagnosing performance issues in Truffle language implementations.

Features

This plugin includes expert-level skills for:

  • Analyzing Compiler Graphs - Understanding IGV dumps and compilation patterns
  • Detecting Deoptimizations - Identifying and fixing runtime deoptimizations
  • Detecting Performance Warnings - Spotting compiler and runtime warnings
  • Establishing Benchmark Baselines - Creating reliable performance benchmarks
  • Fetching Truffle Documentation - Accessing up-to-date Truffle API docs
  • Optimization Workflow Orchestrator - Optimization loop: baseline, analyze & plan, implement & validate, refresh baseline, update lessons learned, clean up, commit
  • Profiling Memory Allocations - Tracking allocation patterns and memory usage
  • Profiling with CPU Sampler - Collecting and analyzing CPU profiles
  • Tracing Compilation Events - Monitoring JIT compilation activity
  • Tracing Execution Counts - Measuring node execution frequencies
  • Tracing Inlining Decisions - Understanding inlining behavior

Installation

Prerequisites

  • Claude Code installed and authenticated (installation guide)
  • Claude Code version 1.0.33 or later (run claude --version to check)
  • uv installed (required for the MCP AWFY benchmark server)

Install the Plugin

  1. Clone or download this repository to your local machine

  2. Load the plugin when starting Claude Code:

    claude --plugin-dir /path/to/cc-truffle-performance-plugin
    
  3. Verify installation by running /help in Claude Code - you should see the plugin's commands listed

Usage

Once installed, Claude will automatically use the appropriate skills based on your Truffle performance analysis needs. The skills are model-invoked, meaning Claude selects and applies them contextually when analyzing your code.

Example Scenarios

  • Ask Claude to analyze compiler graphs from IGV dumps
  • Request help debugging deoptimizations in your Truffle language
  • Get assistance profiling memory allocations
  • Have Claude generate performance reports for your benchmarks

Plugin Structure

cc-truffle-performance-plugin/
├── .claude-plugin/
│   └── plugin.json         # Plugin manifest
├── agents/                 # Sub-agent definitions
│   ├── baseline-establisher.md
│   ├── compiler-graph-analyst.md
│   └── hypothesis-validator.md
└── skills/                 # Skill definitions
    ├── detecting-deoptimizations/
    ├── detecting-performance-warnings/
    ├── fetching-truffle-documentation/
    ├── optimization-workflow-orchestrator/
    ├── profiling-memory-allocations/
    ├── profiling-with-cpu-sampler/
    ├── tracing-compilation-events/
    ├── tracing-execution-counts/
    └── tracing-inlining-decisions/

Development

To modify or extend this plugin:

  1. Edit the skill definitions in the skills/ directory
  2. Update the plugin manifest in .claude-plugin/plugin.json if needed
  3. Restart Claude Code to pick up changes
Skills Info
Original Name:deep-performance-investigationAuthor:antonykamp