yumedzi

10 mods across 1 repository, 2 stars between them.

yumedzi/yumedzi_claude_plugins

Instructions file

Claude Code instructions for yumedzi/yumedzi_claude_plugins, covering claude.md, what this repo is, repo structure, agent conventions (plugins/smart-agents/agents/.md) and licensing note.

2 20d ago A 1,089 tokens

cheap-explore

03

yumedzi/yumedzi_claude_plugins

Plugin Claude Code

Zero-cost enforcement that the built-in Explore agent stays on a cheap model tier. Explore carries no model pin, so a dispatch can silently run at the full session-model rate, or be explicitly set to opus for a plain file sweep. A PreToolUse hook denies any Explore dispatch whose model isn't haiku or sonnet and tells…

2 20d ago A tokens not measured

PreToolUse

04

yumedzi/yumedzi_claude_plugins

Hook

Runs before the agent uses a tool for Agent tool calls, executing require-model-pin.sh via bash. From yumedzi/yumedzi_claude_plugins.

2 20d ago A tokens not measured

context-guard

05

yumedzi/yumedzi_claude_plugins

Plugin Claude Code

Zero-cost context-usage warning. A UserPromptSubmit hook reads the session transcript, infers the context window size from the active model, and shows a one-time systemMessage warning at 35% and again at 50% usage. Never sends anything to the model — the warning never appears in the transcript and costs no tokens.

2 20d ago A tokens not measured

smart-agents

06

yumedzi/yumedzi_claude_plugins

Plugin Claude Code

Four model-pinned delegation agents — scout (Haiku recon), researcher (Sonnet web briefs), worker (Sonnet code changes), deep-thinker (Opus reasoning). Cuts costs by routing to the cheapest model that can do the job. SessionStart hook injects delegation discipline every session; deep-thinker dispatch asks for…

2 20d ago A tokens not measured

deep-thinker

07

yumedzi/yumedzi_claude_plugins

Agent

Heavy reasoning — architecture tradeoffs, root-cause analysis, risk assessment. Read-only. Dispatch when quick implementation would be premature, the problem has no obvious cause and spans multiple modules, or a worker has failed twice and reported up (escalation target of the cheap-first cascade).

2 20d ago A 61 tokens

researcher

08

yumedzi/yumedzi_claude_plugins

Agent

Web researcher — searches and synthesizes into a sourced brief. Dispatch for open-ended external questions needing 3+ sources triangulated (library docs, error messages, API references, comparisons). For a single known URL, WebFetch directly instead.

2 20d ago A 51 tokens

scout

09

yumedzi/yumedzi_claude_plugins

Agent

Fast, cheap codebase recon (Haiku). Dispatch when the task names an area but not specific files, or when orienting would take 5+ greps/reads — returns a structured file map with line ranges and key snippets, not full source. Read-only.

2 20d ago A 58 tokens

worker

10

yumedzi/yumedzi_claude_plugins

Agent

Isolated code worker — reads, writes, edits, runs commands. Dispatch for self-contained code changes with full context in the prompt; protects its own context by spawning scout/researcher. Also runs headless via claude --agent worker.

2 20d ago A 52 tokens