Safety and observability plugins for Claude Code
npx claudepluginhub clemux/claude-code-pluginsBlocks dangerous git commands (force-push, amend, reset --hard, etc.) for all agents
Logs token usage for every subagent (Task tool) call to ~/.claude/subagent-metrics.jsonl
Run tests with coverage and baseline tracking (pytest, with JS/TS support planned)
Smart conventional commits with automatic change analysis and splitting
Asks for confirmation when entering plan mode with a non-Opus model
Manage git worktrees using git gtr (create, copy files, navigate)
Diagnose slow pytest suites, profile with pyinstrument, apply targeted fixes, and produce a performance report
Production-ready workflow orchestration with 79 focused plugins, 184 specialized agents, and 150 skills - optimized for granular installation and minimal token usage
Claude Code marketplace entries for the plugin-safe Antigravity Awesome Skills library and its compatible editorial bundles.
Directory of popular Claude Code extensions including development tools, productivity plugins, and MCP integrations
These plugins are just me playing around with Claude Code's plugin system, and have been iteratively developed by prompting Claude Code to improve them each time the behavior didn't match what I had in mind.
There are probably much better-designed plugins that do the same things with more features; if you know of any, please open an issue and I will happily mention them in this README as better alternatives:
A collection of community plugins for Claude Code that enhance safety, monitoring, and developer experience.
/plugin marketplace add clemux/claude-code-plugins
# Install the safer-git plugin
/plugin install safer-git@clemux-claude-code-plugins
# Install the subagent-metrics plugin
/plugin install subagent-metrics@clemux-claude-code-plugins
# Install the commit plugin
/plugin install commit@clemux-claude-code-plugins
# Install the test-runner plugin
/plugin install test-runner@clemux-claude-code-plugins
# Install the worktree plugin
/plugin install worktree@clemux-claude-code-plugins
Blocks dangerous git commands for all agents (main + subagents), preventing accidental data loss and history rewrites. It's called "safer" rather than "safe" because it reduces risk but cannot guarantee complete protection — an LLM agent can find indirect execution paths (scripts, eval, non-Bash runtimes) that bypass command-level hooks. See known bypass vectors and limitations.
Protects against:
git push --force, --force-with-lease)git commit --amend, git rebase -i, git filter-branch)git reset --hard, git checkout ., git clean -f)git commit --no-verify)git stash drop, git stash clear)git config alias.*)How it works: Registers a PreToolUse hook on the Bash tool that intercepts commands, splits compound statements, and regex-matches each segment against a blocklist. Denies execution with descriptive error messages when dangerous patterns are detected.
Dependencies: jq, grep -P (PCRE support)
Logs token usage for every subagent (Task tool) call, enabling cost attribution and skill optimization.
Features:
~/.claude/subagent-metrics.jsonlflock for concurrency safetyLog fields: timestamp, session ID, model, subagent type, triggering skill, description, total tokens, duration, loaded skills
How it works: Registers a PostToolUse hook on the Task tool that extracts token usage from result text, parses skill tags from descriptions, and appends structured JSON logs.
Dependencies: jq, flock
Smart conventional commits with automatic change analysis and splitting via the /commit command.
Features:
auto mode to skip)Usage: /commit, /commit push, /commit auto
How it works: Uses a heavy-skill/thin-subagent architecture — Opus does all analysis, splitting, and planning in the skill, while a dedicated Haiku subagent executes git commands. The custom commit subagent type enables proper attribution in subagent-metrics logs.
Runs pytest with coverage and baseline tracking via the /tests command.
Features:
.tests-baseline.jsonUsage: /tests, /tests tests/test_auth.py, /tests -k "test_login"