Professional workflow plugins for Claude Code — make Claude apply your project's actual linting rules, commit conventions, and testing standards, not generic defaults. Covers Python, shell, Perl, CI/CD, AI tooling, and agent orchestration.

# Add the marketplace (one-time setup, ~10 seconds, no restart required)
/plugin marketplace add Jamie-BitFlight/claude_skills

# Install a plugin (use the install name from the tables below)
/plugin install dh@jamie-bitflight-skills
/plugin install python-engineering@jamie-bitflight-skills

Start a new session and ask Claude to perform a task the plugin handles (for example, build a CLI with Typer after installing python-engineering). Claude will apply the plugin's conventions rather than generic defaults.

Comprehensive frameworks with multiple skills, commands, and specialized agents.

Focused plugins that teach Claude specific conventions or tools without heavy workflows.

The SAM (Stateless Agent Methodology) pipeline in a single plugin. Every feature request moves through seven stages that each produce a file artifact: Discovery, Planning (with RT-ICA information completeness analysis), Context Integration, Task Decomposition, Execution, Forensic Review, and Final Verification.

Language plugins like python3-development plug into the harness by telling it which specialist agents to use for each task. Without a language plugin, the harness falls back to general-purpose agents.

Install name: dh

Skills include: /dh:add-new-feature, /dh:implement-feature, /dh:complete-implementation, /dh:groom-milestone, /dh:work-milestone, /dh:dispatch, /dh:backlog, and more.

Agents include: @dh:swarm-task-planner, @dh:feature-researcher, @dh:codebase-analyzer, @dh:feature-verifier, @dh:doc-drift-auditor, and more.

MCP servers: Backlog server (GitHub Issues sync, artifact management, dispatch orchestration), sequential-thinking server.

Python specialist that composes with development-harness. Install both for the full Python development pipeline.

Skills include: Python 3.11+ patterns, Typer/Rich CLI development, pytest workflows, type checking, linting resolution, PEP 723 inline script metadata, pyproject.toml configuration, and more.

Agents include:

• @python3-development:python-cli-architect — implements Python CLI features end-to-end
• @python3-development:python-cli-design-spec — produces architecture specs
• @python3-development:python-pytest-architect — writes pytest test suites
• @python3-development:code-reviewer — code review with Python idiom awareness
• @python3-development:semantic-code-search — semantic search over Python codebases

MCP servers: cocoindex-code semantic code search, sequential-thinking server.

Opinionated Python 3.11+ engineering system that establishes strong defaults and routes to specialist skills. Composes with development-harness for the full pipeline, or works standalone.

Skills include: One automatic router (python3-core) plus specialist skills for CLI (python3-cli), testing (python3-testing, python3-tdd), typing (python3-typing), web (python3-web), data (python3-data), tooling (python3-tools), and constrained environments (python3-stdlib-only). Manual entrypoints: /python-engineering:orchestrate, /python-engineering:review, /python-engineering:lint, /python-engineering:cleanup, /python-engineering:debug.

Agents include:

• @python-engineering:python-cli-architect — implements Python CLI features end-to-end
• @python-engineering:python-cli-design-spec — produces architecture specs for CLIs
• @python-engineering:python-pytest-architect — writes pytest test suites
• @python-engineering:code-reviewer — code review with Python idiom awareness
• @python-engineering:semantic-code-search — semantic search over Python codebases

Toolkit for building, refactoring, and validating plugins. Claude won't drift from current schema when creating agents or skills — the plugin loads the authoritative reference documentation automatically.

Skills include: /plugin-creator:skill-creator, /plugin-creator:agent-creator, /plugin-creator:plugin-lifecycle, /plugin-creator:refactor-plugin, /plugin-creator:refactor-skill, /plugin-creator:claude-skills-overview-2026, /plugin-creator:claude-plugins-reference-2026, /plugin-creator:hooks-guide, and more.

Agents include: refactor-planner, refactor-executor, refactor-validator, subagent-refactorer (applies Anthropic prompt engineering best practices), skill-auditor, skill-content-updater, ai-doc-optimizer, plugin-assessor.

Scripts include: create_plugin.py (interactive scaffolding), auto_sync_manifests.py (pre-commit hook that bumps versions and syncs component arrays automatically).

Validation: Integrates with skilllint (uvx skilllint@latest check <path>) for frontmatter validation, skill complexity checking, and auto-fix of common errors.

Installs three PreToolUse hooks that run before every Read, Grep, and Bash call:

• Read/Grep hook — warns when Claude reads a source file it is not about to edit, prompting delegation instead
• Diagnostic command gate — blocks linter/type-checker/test commands from running directly in the orchestrator context window
• Bash misuse prevention — catches other patterns where the orchestrator should delegate rather than act

These hooks structurally prevent the investigation escalation anti-pattern: the progressive cycle of reading files that justifies reading more files until Claude implements the task itself instead of delegating.

Two MCP servers surface transcript analysis directly in your session:

• kaizen-duckdb — DuckDB read-only access to the session transcript database for SQL-based process mining
• kaizen-analysis — analysis server for anti-pattern detection, sentiment signals, and improvement recommendations

Run after sessions to identify what went wrong, what tooling Claude was missing, and where repeated mistakes occurred.

Covers the full FastMCP 3.x development surface: provider/transform architecture (CodeMode, Tool Search, server-level transforms), component versioning, session state, MultiAuth authorization, PropelAuth integration, Pydantic validation, async patterns, STDIO/HTTP transports, nginx reverse proxy deployment, background tasks, Prefab Apps UI, security patterns, client SDK usage, testing, and migration from FastMCP v2.

Includes a live FastMCP reference MCP server that provides search_docs, scaffold_server, validate_server, and version_check tools.

A browser-based control panel for Claude Code. When you are running multiple sessions simultaneously — across projects, worktrees, or machines — dot-dash gives you a single place to watch what each session is doing and to interact with it.

What it does:

• Live transcript view — streams JSONL transcript events from each session to the browser in real time via WebSocket, so you can read Claude's reasoning and tool calls as they happen
• Session list — shows all active and recently stopped sessions with project name, working directory, PID, and last-event timestamp
• Prompt injection — lets you queue a message in the browser that will be prepended to the next prompt the user submits in a specific session, without interrupting the current turn
• Session termination — send a stop signal to any session from the dashboard

How it works:

Three hooks wire each Claude Code session into the server automatically:

• SessionStart — registers the session (ID, CWD, PID) with the local server
• SessionEnd — marks the session stopped and cleans up
• UserPromptSubmit — checks the injection queue before every prompt; if a message is queued, prepends it transparently

The server runs locally at http://127.0.0.1:7765 by default. A React/TypeScript frontend (built with Vite) is served from the same process. Authentication uses a token stored at ~/.claude/dot-dash/token (generated on first run, file-permissions 600).

Starting the server:

cd plugins/dot-dash/server
npm install
npm run build
node dist/main.js

Or use the bundled script:

bash plugins/dot-dash/scripts/start-server.sh

Then open http://127.0.0.1:7765 in a browser. The server must be running before Claude Code sessions start for those sessions to appear in the dashboard.

Configuration:

Set DOT_DASH_PORT in your environment to use a port other than 7765:

DOT_DASH_PORT=8080 node dist/main.js

Plugins contain:

• Skills — Knowledge and workflows that guide Claude's behavior
• Commands — Slash commands you can invoke directly (like /dh:add-new-feature)
• Agents — Specialized sub-agents for complex tasks (like code review or architecture design)
• Hooks — Automation that runs at specific lifecycle events
• MCP Servers — Additional tools Claude can call (backlog management, code search, file reading, etc.)

Once installed, plugins work automatically. Claude knows when to apply them based on your project context.

plugins/plugin-name/
├── .claude-plugin/
│   └── plugin.json       # Plugin manifest
├── skills/               # What Claude learns
├── commands/             # Slash commands you can use
├── agents/               # Specialized sub-agents
├── hooks/                # hooks.json and hook scripts
└── README.md             # Documentation

# Option 1: Load specific plugins for this session
claude --plugin-dir ./plugins/python-engineering --plugin-dir ./plugins/holistic-linting

# Option 2: Add local marketplace for persistent enable/disable
/plugin marketplace add ./.claude-plugin/marketplace.json

# Install plugins you need (--scope local keeps it gitignored)
/plugin install python-engineering@jamie-bitflight-skills --scope local

# Disable when not needed
/plugin disable python-engineering@jamie-bitflight-skills

# Re-enable when needed
/plugin enable python-engineering@jamie-bitflight-skills

Plugin commands not found after install?

Restart Claude Code to reload commands. Verify the plugin installed:

/plugin list

Plugin not influencing Claude's behavior?

Check that the plugin is enabled and not scoped to a different project:

/plugin list --all

Want to test a plugin before committing to it?

Use --scope local when installing. This keeps the plugin active only in your current project and gitignored.

Something broken after an update?

Reinstall the specific plugin (use the install name from the tables above):

/plugin install <install-name>@jamie-bitflight-skills

/plugin marketplace add Jamie-BitFlight/claude_skills
/plugin install dh@jamie-bitflight-skills

Every plugin works immediately after install — no restart, no configuration.

See CONTRIBUTING.md for detailed guidelines on adding, removing, or updating plugins.

Quick overview:

1. Fork the repository
2. Create your plugin using /plugin-creator or manually
3. Update .claude-plugin/marketplace.json
4. Validate with uvx skilllint@latest check
5. Test locally before submitting PR
6. Submit pull request with description

• Claude Code v2.0 or later
• Individual plugins may have additional requirements (see plugin READMEs)

MIT License — see individual plugins for specifics.