YeonwooSung/my-claude-code-mods

My Claude Code Mods – what it is

A collection of plugins (called “mods”) for Claude Code, Anthropic’s code‑assistant environment. Each mod adds a small UI or diagnostic feature that helps developers understand and control what Claude is doing during a coding session.

The individual mods

Mod	What it does (plain language)
agent‑profiler	Shows where the session’s time is spent – waiting for the model, running tools, invoking sub‑agents, or re‑reading code. It can also export OpenTelemetry traces that you can view in Jaeger or Tempo.
context‑inspector	Analyzes how Claude uses the prompt window: what text fills the window, how often it re‑reads, whether it produces near‑duplicate output, and what parts get compacted. It then gives advice on reducing wasted context.
time‑machine	Records every tool call on a branching timeline, taking “shadow‑git” snapshots of the file system after each step. You can rewind to any step, see which edit broke a test, restore files, create branches, compare runs, or replay steps.
loop‑detective	Watches for repeated Bash failures and edit‑retry cycles. When it spots a loop it presents evidence, prompts for investigation, and shows a live case pane with a bounded history.
pin‑memo	Lets you pin notes and clickable TODOs above the input box. The notes are automatically restored when you reopen the same folder or session.
model‑effort‑band	Draws a coloured band above the prompt that indicates which model is selected and how much main‑thread effort the model is using, giving a quick visual cue of workload.

How to get them

Add the Claude Code Mods marketplace (see the official docs linked in the README).
Install any of the mods you want via the marketplace – the README for each mod contains the exact command.
When running Claude Code, load a mod with --plugin-dir (or enable it through the UI). Other installed plugins can be disabled for clean benchmarking.

Development & testing

Node version – requires Node 22.19.0 or newer.
Dependencies – locked in package-lock.json.
Test commands:
- npm ci --ignore-scripts – install exact deps.
- npm test – runs core logic tests that do not need a Claude installation.
- CLAUDE_BIN=/path/to/claude npm run test:adapters – runs the full engine suites, including the adapters that talk to Claude.
- npm run typecheck – checks TypeScript types against locally generated engine declarations.
Compatibility matrix in the README shows which Claude Code versions each mod has been verified against (mostly 2.1.292, with some early‑access notes).
UI‑related tests are separate from the core unit suite; they require an interactive run.

Performance benchmarking (as described)

The README suggests a controlled experiment:

Pick a large repository at a fixed commit.
Run the same workload under four configurations: no mods, profiler only, time‑machine only, and all three.
Load mods explicitly with --plugin-dir and disable any others.
Warm‑up, then perform at least five measured runs per configuration, alternating order.
Record wall‑clock time, number of tool calls, model token usage, and average snapshot cost.
Report median and 95th‑percentile values, separating pure model latency from the overhead of snapshots/hooks.

Who might use this

Developers who rely on Claude Code for code generation and want visibility into model latency, context usage, or tool‑call history.
Teams that need reproducibility (time‑machine) or want to catch infinite‑loop bugs in Bash scripts (loop‑detective).
Anyone who likes to keep quick notes attached to a coding session (pin‑memo).

Bottom line

My Claude Code Mods is a small, focused toolbox that extends Claude Code with diagnostics, history‑tracking, and UI helpers. The repository provides the source, a Node‑based test harness, and clear instructions for installing each mod via the official Claude Code marketplace. It is a genuine, AI‑related software project aimed at improving the developer experience when using a large‑language‑model coding assistant.