English | 한국어
Claude Code plugins by haroom.
A small collection of Claude Code plugins that target specific pain points when using Claude Code for real work: parallel sessions clobbering each other, AI reviewing its own code, design docs that AI cannot parse. Each plugin lives in its own repo; this marketplace aggregates them so one marketplace add covers all four.
| If this sounds like you... | Try |
|---|---|
| "I run multiple Claude Code sessions and they trip over each other" | ddaro |
| "I want a 5-angle code review before a risky PR" | prism |
| "This design doc will be read by AI agents too, I need it airtight" | triad |
| "I want Claude and Codex to cross-review a file until it converges" | mangchi |
Details per plugin below.
/plugin marketplace add https://github.com/minwoo-data/haroom_plugins.git
/plugin install ddaro
/plugin install prism
/plugin install triad
/plugin install mangchi
Required after every install and update.
/plugin update
One command refreshes every installed haroom plugin from this marketplace.
Pain: two Claude Code sessions editing the same repo overwrite each other's work; sessions die without saving context.
What it does: creates one git worktree per session (disposable folder + branch), auto-pushes every commit, writes a plain-text context snapshot to disk so a crash never loses the picture, and refuses to merge if a file on main would silently disappear.
Use when: running 2+ Claude Code sessions, doing risky refactors, keeping long-running experimental branches, or working on crash-prone machines.
Pain: a single reviewer misses what a different angle would have caught. LLM self-review has predictable blind spots.
What it does: five specialized review agents look at the target in parallel (conflict detection / improvement / devil's advocate / code review / robustness), then a verifier pass cross-checks findings only one agent flagged so false positives do not waste your time.
Use when: before a major architectural decision, before merging an uncertain PR, or auditing a skill / design doc / workflow.
Pain: you ship a design doc and later discover AI agents cannot parse it, the architecture ages poorly, or new users cannot figure out what it says.
What it does: three independent agents review one document through fixed lenses (LLM clarity / architectural longevity / end-user comprehension) across rounds until they agree or hit the cap. Consensus is written to updated.md without touching the original.
Use when: writing project instruction docs (CLAUDE.md, agent prompts), ADRs, public docs, or finalizing a PR justification.
Pain: one model reviewing its own code reinforces its own blind spots; the same bug survives multiple review passes.
What it does: alternates Claude-as-editor with Codex-as-reviewer across five rotating axes (correctness / security / readability / performance / design), catching each model's blind spots with the other until the code converges.
Use when: hardening production code before release, reviewing prompt-injection / security defense code, or using Claude Max tokens to drive external cross-model review without running Codex by hand.
Each plugin is pulled in as a git submodule at plugins/<name>/, pointing at its own GitHub repo. When a plugin ships a new release, this aggregator's submodule pointer is bumped automatically every day at 06:00 UTC by .github/workflows/daily-submodule-update.yml. /plugin update then picks up the new version on the next refresh.
MIT - see LICENSE. Each plugin retains its original LICENSE file preserving provenance.
haroom - github.com/minwoo-data