Claude Code Plugins

Developer workflow, productivity, and writing plugins for Claude Code.

See: Claude Code: Plugin Marketplaces

Plugins

Plugin	Description
dev-workflow	Workflows for common development tasks.
writing-tools	Writing and editing tools for content creation.

Installation (Claude Code)

Add the marketplace once, then install the plugins you want:

claude plugin marketplace add noahlz/claude-plugins
claude plugin install dev-workflow@noahlz.github.io
claude plugin install writing-tools@noahlz.github.io

Prerequisites

• Node.js 22+ (required for dev-workflow only)

Installation (Claude.ai Web)

Several writing-oriented skills are available standalone for use on claude.ai. Accounts must be Pro, Max, Team, or Enterprise plan with code execution enabled.

Available web-packaged skills:

• de-llm
• craft-linkedin-post — bundles the linkedin-reviewer agent
• tighten-for-llms

See writing-tools and dev-workflow sections below for descriptions.

To install:

1. Download zip from the latest release page
2. In Claude.ai, open Settings → Capabilities and choose Upload skill.
3. Drag the skill ZIP file to Upload skill. The skill will now be available in your chats.

dev-workflow

Workflows for common development tasks.

Skills

Skill	Description	Slash Command
run-tests	Run tests, analyze failures, and propose fixes via sub-agents.	/run-tests
commit-with-costs	Create git commits with Claude Code attribution and session cost metrics.	/commit-with-costs
merge-with-costs	Merge sub-agent worktree branches, resolve any conflicts, and record the project session's cost in the merge commit.	/merge-with-costs
preview-commit-message	Draft a commit message from staged changes without committing.	/preview-commit-message
commit-only	Commit staged changes with an AI-generated message. No cost metrics.	/commit-only
view-cost-metrics	Display current session cost metrics without creating a commit.	/view-cost-metrics
check-claude-changelog	Check Claude Code changelog for updates since your last commit, with relevance scoring.	/check-claude-changelog [since <version>]
update-changelog	Inspect commits since last release and compose changelog entries matching your project's format.	/update-changelog
tighten-for-llms	Make skill, agent, reference, rule, or doc files more concise and LLM-friendly.	/tighten-for-llms [file-path | skill-name | agent-name | docs | readme | rules]
zero-shot	Experimental. Aggressively distill LLM-facing prose: skill files → 1–3 imperative sentences, instruction files → imperative bullets per section.	/zero-shot [apply] [file-path | skill-name | agent-name | rules]

Agents

Agent	Description
broken-build-analyzer	Analyzes build/compilation failures and provides diagnosis suitable for creating a plan to fix the build. Used by run-tests.
failed-test-analyzer	Analyzes test failures and provides diagnosis suitable for creating a plan to fix the tests. Used by run-tests.
react-code-reviewer	Reviews React/TypeScript code changes for quality, maintainability, and best practices. Reports findings by priority with concrete fixes.
test-quality-reviewer	Reviews written or modified tests for assertion quality, mock validity, and clean test structure. Reports findings by severity.

writing-tools

Writing and editing tools for content creation.

Skills

Skill	Description	Slash Command
craft-linkedin-post	Draft, review, and iterate a LinkedIn post from an outline. Uses the linkedin-reviewer agent.	/craft-linkedin-post
de-llm	Post-process prose to strip AI-telltale patterns: puffery, inline meta-commentary, unsourced evaluative claims. Derived from Wikipedia: Signs of AI writing.	/de-llm [file-path]

Agents

Agent	Description
linkedin-reviewer	Reviews a LinkedIn draft post for hook strength, clarity, mobile structure, authenticity, and engagement potential.

Development

Local Development Setup

When developing these plugins, you can install them from local source:

claude plugin marketplace add ./

Updating

Claude Code caches installed plugins as full copies in ~/.claude/plugins/cache/<marketplace>/<plugin>/<version>/. These are regular files, not symlinks — editing your source directory does not change what Claude Code reads.

/reload-plugins reloads from this version-keyed cache. If you modify plugin files without bumping the version in marketplace.json, /reload-plugins will reload the stale cached copy and not pick up your source changes.

To pick up source changes without bumping the version, run ./force-reinstall.sh:

./force-reinstall.sh

This script clears the cache directories, uninstalls, and reinstalls from local source.

Use this when:

• You've modified SKILL.md, scripts, agents, or other plugin files without bumping the version
• /reload-plugins is not reflecting your changes

Plugin Root Resolution

Claude Code substitutes two variables in SKILL.md content at load time:

• ${CLAUDE_PLUGIN_ROOT} – the plugin's installed directory. Skills use it for shared code in lib/ and shared docs in references/.
• ${CLAUDE_SKILL_DIR} – the skill's own directory (added in Claude Code 2.1.69). Skills use it for their own scripts/.

Minimum required version: Claude Code 2.1.69

Releasing

release-it drives releases, configured in .release-it.json.

Never edit version numbers by hand. The @release-it/bumper plugin writes all of them: package.json, package-lock.json, and .claude-plugin/marketplace.json (both metadata.version and each plugin's version).

Step	Command	Notes
1. Write the changelog	/update-changelog	Composes CHANGELOG.md entries from commits since the last tag
2. Preview	npm run release:dry-run	Shows the version bump and files it would touch
3. Release	npm run release	Bumps versions, commits Version X.Y.Z, tags vX.Y.Z
4. Push	git push --follow-tags	.release-it.json sets git.push: false, so this is manual

The changelog entry must land before the tag. release-web-skills.yml extracts the GitHub release notes by matching a # X.Y.Z heading in CHANGELOG.md — a single #, version only, no v prefix. A missing or misspelled heading produces an empty release body.

Pushing a v* tag builds the claude.ai-uploadable ZIPs listed in scripts/build-web-skills/manifest.js and attaches them to the GitHub release. To dry-run that build without cutting a tag, dispatch the workflow manually — it uploads the zips as workflow artifacts instead.

Version choice: a new skill, agent, or plugin is a minor bump; fixes and refinements to existing ones are a patch.

Testing

Scripts used by skills are tested using Node.js with the native node:test module.

Run tests with this plugin's dev-workflow:run-tests skill ("eat your own dogfood") or:

npm test

See test/CLAUDE.md for testing methodology, patterns, and examples.

Authors

@noahlz
Claude

License

MIT License (MIT)