Claude Skills Collection

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, and AI tooling.

What Problem Does This Solve?

Without plugins	With plugins
Claude gives generic Python advice	Claude applies Python 3.11+, Typer, Rich, httpx conventions specific to your stack
Claude says "done" before linters pass	holistic-linting enforces root-cause fixes before any task completes
Claude speculates and hallucinates	hallucination-detector blocks completion on ungrounded claims
Claude jumps to solutions without investigating	verification-gate forces evidence gathering before action
Session transcripts disappear with no learning	agentskill-kaizen mines transcripts for anti-patterns and generates skill patches
Commit messages are inconsistent	conventional-commits enforces feat/fix/chore format for semantic versioning
Claude reads source files when it should delegate	orchestrator-discipline hooks block investigation escalation at the tool level

Quick Start

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

# Install a plugin
/plugin install plugin-name@jamie-bitflight-skills

Verify Installation

Start a new session and ask Claude to perform a task the plugin handles (for example, build a CLI with Typer after installing python3-development). Claude will apply the plugin's conventions rather than generic defaults.

Available Plugins

Full-Featured Development Systems

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

Plugin	What It Does
development-harness	Language-agnostic SAM 7-stage pipeline (Discovery → Planning → Context → Decomposition → Execution → Review → Verification) with backlog management, milestone dispatch, and kage-bunshin parallel sessions. 40 skills, 19 agents, MCP backlog server.
python3-development	Python 3.11+ specialist with 34 skills, 5 agents, and TDD workflows. Covers Typer/Rich CLI development, pytest test suites, code review, type checking, and PEP 723 inline script metadata. MCP semantic code search included.
bash-development	Write robust Bash 5.1+ scripts with modern patterns, error handling, POSIX portability, and specialized agents for development and auditing
perl-development	Build production-quality Perl 5.30+ scripts with modern practices, CPAN ecosystem integration, comprehensive testing, and CLI architecture
python-engineering	Opinionated Python 3.11+ engineering system with 39 skills and 5 agents. Establishes strong defaults (SOLID, typing policy, testing standards, code smell detection) and routes to specialist skills for TDD, CLI (Typer/Rich), web, data/science, and constrained environments.
plugin-creator	Complete toolkit for creating, refactoring, and validating Claude Code plugins with 40 skills, 8 specialized agents, automated version bumping, and `skilllint` integration
uv	Expert guidance for Astral's uv — the fast Python package manager that replaces pip, poetry, pyenv, and virtualenv with modern lockfiles
clang-format	Stop clang-format from reformatting your existing C/C++ style — analyzes your code's patterns first and shows the impact before applying any changes (install name: `clang-format-configuration`)
holistic-linting	Automatic code quality enforcement — Claude won't say "done" until code passes all configured linters with root-cause fixing. Covers ruff, mypy, and bandit.
summarizer	Faithful information summarization with anti-hallucination methodology, structured output templates, and autonomous agents for file, URL, and image summarization
agentskill-kaizen	Analyze Claude Code session transcripts to find inefficiencies, anti-patterns, and repeated mistakes with DuckDB process-mining and live sentiment dashboard. Two MCP servers included.
dasel	Query, transform, and convert structured data files (JSON, YAML, TOML, XML, CSV, HCL, INI) using dasel v3 with exploration agents
the-rewrite-room	Documentation workflow router — routes tasks like drift audits, doc sync, prompt optimization, and summarization to canonical workflows with validation (install name: `rwr`). Includes MCP file-reader server.
process-siren	Converts bullet steps, ASCII art, markdown tables, and prose workflows into Mermaid diagrams for AI-facing documents, with process quality methodology for improving ambiguous or incomplete processes before conversion
fastmcp-creator	Build production-ready MCP servers with FastMCP 3.x — covers provider/transform architecture, authorization, session state, async patterns, STDIO/HTTP transports, and deployment. Includes a live FastMCP reference MCP server.
dot-dash	Real-time browser dashboard for monitoring every active Claude Code session — live transcript streaming, prompt injection, and session termination from a single React UI. Node.js/Hono backend with WebSocket updates and token-based authentication.

Lightweight Knowledge Clip-Ins

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

Python and Package Management

Plugin	What It Does
litellm	Call any LLM API (OpenAI/Anthropic/local) from Python with unified interface and retry logic
llamafile	Run local GGUF models with OpenAI-compatible API for offline/air-gapped inference
xdg-base-directory	Store config and data files in XDG-compliant directories using platformdirs — so your tool works on Linux, macOS, and Windows without hardcoded paths breaking on other systems

Git and CI/CD

Plugin	What It Does
conventional-commits	Write consistent commit messages (feat/fix/chore) for semantic versioning and changelog generation
commitlint	Configure and validate commit messages against commitlint rules — CI rejects non-conforming commits before they merge
gitlab-skill	Write GitLab CI pipelines and GLFM documentation with local testing via gitlab-ci-local before pushing

Better Claude Behavior

Plugin	What It Does
agent-orchestration	Structures delegation prompts with world-building context (WHERE, WHAT, WHY) while preserving agent autonomy on implementation decisions
verification-gate	Forces Claude to verify its hypothesis matches its target before executing writes or edits — blocks correct diagnoses from producing wrong implementations
hallucination-detector	Blocks task completion when Claude speculates or makes ungrounded claims, forcing evidence-first rewrites
scientific-method	Structures hypothesis-driven debugging and investigation with experiment protocols and evidence-first methodology
brainstorming-skill	Significantly improves brainstorming with 30+ research-validated prompt patterns across 14 categories
frustration-analyzer	Finds the strongest user reaction to an AI instruction-following failure in a session, reconstructs the triggering assistant output, and renders a shareable terminal-style PNG rage receipt. 3 agents, MCP server.
orchestrator-discipline	Enforces context window discipline via PreToolUse hooks — blocks source-file reads without edits and blocks diagnostic commands that should be delegated to agents

Architecture

Plugin	What It Does
twelve-factor-app	Apply twelve-factor app methodology (15 principles including 3 modern extensions) to your projects for portable, scalable, cloud-native architecture

Plugin Details

development-harness

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.

Skills include: /dh:add-new-feature, /dh:implement-feature, /dh:complete-implementation, /dh:groom-milestone, /dh:work-milestone, /dh:dispatch, /dh:backlog, and 33 more.

Agents include: @dh:swarm-task-planner, @dh:feature-researcher, @dh:codebase-analyzer, @dh:feature-verifier, @dh:doc-drift-auditor, and 14 more.

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

python3-development

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.

python-engineering

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

plugin-creator

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), contextual-ai-documentation-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.

orchestrator-discipline

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.

agentskill-kaizen

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.

fastmcp-creator

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.

dot-dash

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

How Plugins Work

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.

Plugin Structure

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

Local Development

# Option 1: Load specific plugins for this session
claude --plugin-dir ./plugins/python3-development --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 python3-development@jamie-bitflight-skills --scope local

# Disable when not needed
/plugin disable python3-development@jamie-bitflight-skills

# Re-enable when needed
/plugin enable python3-development@jamie-bitflight-skills

Troubleshooting

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:

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

Get Started Now

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

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

Contributing

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

Quick overview:

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

Requirements

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

License

MIT License — see individual plugins for specifics.

Name	create-feature-task
Description	This skill should be used when the user asks to "create a feature task", "set up development tracking", "plan a feature implementation", or needs to structure a new feature development with proper tracking and phases.

create-feature-task

SKILL.md