# Claude skills marketplace: four plugins, one shell installer, and savings it estimates

> mhattingpete/claude-skills-marketplace packages Claude Code plugins, agents and a local Python execution runtime that the author credits with 90 to 99 percent token reduction on bulk file work. The claims are arithmetic in the README rather than measurements, and the git skill will push your branches without asking.

**mhattingpete/claude-skills-marketplace** — Claude Code Skills for software engineering workflows - Git automation, testing, and code review

- Repository: https://github.com/mhattingpete/claude-skills-marketplace
- Stars: 680 · Forks: 97
- Language: HTML
- License: Apache-2.0
- Published: 2026-09-10 · Updated: 2026-09-10 · Language: en
- Canonical page: https://hysenlabs.com/projects/mhattingpete-claude-skills-marketplace

## Skills trigger themselves, slash commands do not

The distinction the marketplace is built on is between two extension mechanisms in Claude Code. A slash command waits for you to type it. A skill does not: it is a model-invoked capability that Claude activates by itself based on context and on the skill's own description, which means the description field is doing the routing work rather than documentation. Each skill is a single SKILL.md file with YAML frontmatter carrying a name, a description and metadata, detailed instructions underneath, and optional supporting files such as scripts, templates and references. Agents are the other half. Each is an AGENT.md file with frontmatter that includes model selection, plus specialised instructions and decision-making frameworks for its domain, and it runs as an independent instance with its own context. The two compose in both directions, since a skill can decide when to invoke an agent and an agent can use skills while it works.

## The token saving is the author's arithmetic, and the README shows it

The headline number is 90 percent or more token savings, with the body of the documentation widening that to a 90 to 99 percent reduction for bulk operations. Two figures support it. The first is a rule of thumb: process 100 files with 1K tokens instead of 100K, which is a 99 percent reduction on its face. The second is a worked example, a rename of one function to another across all Python files, priced two ways. Without execution mode it is given as roughly 25,000 tokens for reading and editing 50 files; with execution it is roughly 600 tokens for a script and a summary, which is a 97.6 percent saving and matches the stated percentage. Both numbers are the author's estimates for illustrative cases rather than measurements on a real repository, which is the distinction to hold on to. The claimed mechanism is the Anthropic code execution pattern: instead of loading code through context, the work runs as local Python with API access.

## Code leaves the context window and runs locally with masking on

Four benefits are claimed for the execution runtime, and two of them are about state rather than tokens. Local execution replaces repeated API round-trips with a script that runs on your machine, which is why bulk work is faster as well as cheaper. Stateful workflows let a multi-step refactoring be resumed across sessions instead of restarting from a fresh context. And the runtime is described as auto-secure, with PII and secret masking plus sandboxed execution, which is the claim a security reviewer should test first rather than accept. Activation is automatic and rule-based: skills are expected to switch to execution mode for bulk operations touching ten or more files, for complex multi-step workflows, for iterative processing, and for codebase-wide analysis. The runtime is laid out as an importable API library, a FastMCP server, a one-command setup script and its own documentation, so the pieces are separable rather than monolithic.

## One setup script finishes the runtime install

Installing the marketplace does not install the runtime, which is the step most people will miss. The plugin add gives you the skills and the agent; the runtime has its own two-minute setup, run as a shell script from inside your plugins directory after the marketplace plugin is installed:

```bash
~/.claude/plugins/marketplaces/mhattingpete-claude-skills/execution-runtime/setup.sh
```

That path is worth reading closely, because it hardcodes both the marketplace directory name derived from the repository and the execution-runtime subdirectory. Rename the marketplace or install from a fork under a different name and the documented path no longer matches, so anyone automating setup should resolve the path rather than copy it. Everything else about the runtime is documented in its own README, and the two modes it offers, an importable API library and a FastMCP server, mean it can be used by other tools rather than only by the skills that ship here.

## Four plugins, installable whole or one at a time

The marketplace manifest at the root exposes the collection, and each of the four sub-plugins can be addressed individually by path. Adding the repository installs everything, and the documentation notes that this brings in the engineering workflow plugin with all of its skills plus the plan-implementer agent. Adding a single sub-path instead installs only that group: engineering workflows for git operations, test fixing, code review implementation and feature planning; visual documentation for HTML documentation with modern UI design, SVG diagrams, flowcharts, dashboards, timelines and architecture diagrams; productivity skills for usage-pattern analysis, code quality audits, project bootstrapping and documentation generation; and code operations for high-precision manipulation such as line-based insertion, bulk refactoring and file analysis. That last plugin has a history worth noting, since it was converted from a separate MCP project into native Claude Code skills.

## The git skill pushes without asking you first

Each published skill declares when it activates and gives example phrasings, which makes the permission surface easy to audit. git-pushing is the one to look at hardest: it stages changes, commits them with conventional commit messages and pushes to the remote, and it activates when you mention pushing changes, committing work or saving to remote, with example triggers like Push these changes, Commit and push to github and Let's save this work. Nothing in that description asks for confirmation first. feature-planning breaks a request into tasks meant to be executed by the plan-implementer agent, with example requests such as adding authentication or building an analytics dashboard. test-fixing works through failing tests with error grouping strategies, and review-implementing turns reviewer comments into tracked work. code-execution is the newest of the set and the one that routes work into the runtime.

## The manifest says HTML and the tree is markdown, shell and Python

A few repository-level details frame how new this is. The primary language recorded for the repository is HTML, which reflects the visual documentation plugin rather than the substance, since the working parts are markdown skill and agent files, a shell installer and a Python execution runtime. The licence is Apache-2.0, there is a changelog, a contributing guide and a project-level CLAUDE.md, and a .claude directory, which means the marketplace ships with agent configuration for the very tool it extends. The last commit was 25 July 2026, roughly two months before this review, and the repository publishes no GitHub releases, so there is no tag to pin and the changelog is the only version history. The project is also listed in the Smithery skills directory under the mhattingpete namespace, which is a second distribution channel with its own rules.

## Conclusion

This marketplace is worth installing if you already live in Claude Code and want file-level bulk work to stop consuming your context, because the execution runtime is the substantive part and the skills are thin wrappers around it. Two things to check first. The token savings are the author's own arithmetic on two examples rather than a benchmark, so measure them on your own repository before reorganising around them. And git-pushing stages, commits and pushes on activation, so decide whether you want an agent with that permission before you add the plugin rather than after.

## FAQ

### What is the Claude skills marketplace by mhattingpete?

It is a curated marketplace of Claude Code plugins for software engineering workflows, holding four plugin groups, a plan-implementer agent and a local Python execution runtime. You add it with /plugin marketplace add mhattingpete/claude-skills-marketplace, or add a single sub-plugin by path, and run the runtime setup script separately.

### Is there a Claude skills marketplace?

Several exist, and this repository is one of them. It installs with a single command in Claude Code, /plugin marketplace add mhattingpete/claude-skills-marketplace, and each of its four plugins can also be added individually by its sub-path when you only want one area of functionality.

### What is a skills marketplace?

A collection of plugins for Claude Code published under one manifest, where each plugin bundles skills, agents and supporting files. Skills are model-invoked and live in SKILL.md files with YAML frontmatter, agents are independent specialised instances described in AGENT.md files with model selection, and the two can call each other.

### What do Claude skills actually do?

They extend Claude Code with capabilities it invokes by itself. Unlike a slash command, which waits for you to type it, a skill activates from context and its own description, and it consists of a SKILL.md file with frontmatter, detailed instructions and optional scripts, templates or references. Agents sit alongside them as independent instances with their own context and model.

## Sources

- [Issues](https://github.com/mhattingpete/claude-skills-marketplace/issues)
- [License: Apache-2.0](https://github.com/mhattingpete/claude-skills-marketplace/blob/main/LICENSE)
- [mhattingpete/claude-skills-marketplace on GitHub](https://github.com/mhattingpete/claude-skills-marketplace)
- [README](https://github.com/mhattingpete/claude-skills-marketplace/blob/main/README.md)

---

Hysen Labs editorial analysis, written from the project's own repository and release notes. Cite the canonical page: https://hysenlabs.com/projects/mhattingpete-claude-skills-marketplace