active-skills

active-skills

A curated set of agent skills, installable as a plugin in both Claude Code and Antigravity. This repository is the source of truth — clone it to author skills. The mlarkin00/claude marketplace mirrors it automatically, so users install everything from that one place and never clone this repo.

The repository root is the plugin. Claude Code reads .claude-plugin/plugin.json; Antigravity reads plugin.json. The two manifests coexist and each carries its own version, so one directory serves both runtimes.

Everything here is skills or skill-authoring tooling. Usage tracking lives in a separate skill-usage plugin, deliberately: keeping plugin machinery out of this repo keeps it a clean place to write skills.

Install

Claude Code — via the mlarkin00-claude marketplace:

/plugin marketplace add mlarkin00/claude
/plugin install active-skills@mlarkin00-claude

Skills are namespaced under the plugin, e.g. active-skills:systematic-debugging.

Antigravity — installs straight from the repository URL:

agy plugin install https://github.com/mlarkin00/active-skills

Authoring

Each skill is a directory under skills/ containing a SKILL.md. That is the whole contract — skills/ must contain nothing but skill directories, because Antigravity installs every entry there as a skill and a loose file becomes a phantom skill in its UI. The eval suite lives in evals/ for that reason.

After adding, removing, or retitling a skill, regenerate the inventory below:

bash scripts/gen-readme.sh

To publish: bump the version in plugin.json and .claude-plugin/plugin.json, then push to main. A sync workflow in mlarkin00/claude mirrors the change into the marketplace and updates its marketplace.json to the new version. The bump is what matters — plugin caches are version-keyed, so a skill change shipped without a version bump will not reach anyone. The sync surfaces a warning when content changes without a bump.

Layout

Path	Purpose
skills/	The skills. Only skill directories belong here.
evals/	Eval suite, kept out of skills/ on purpose.
scripts/gen-readme.sh	Regenerates the inventory below.
sidecars/check-updates/	Antigravity: periodic update check.
tests/	Tests for the update-check sidecar.
.claude-plugin/plugin.json, plugin.json	The two runtime manifests.

Skills

39 skills (auto-generated — do not edit by hand):

• auto-mode — Plan-driven autonomous execution where the agent writes a full plan, batches clarifying questions up front, and executes phase-by-phase with testing and summary logging.
• brainstorming — Use this skill when you need to turn ideas into fully formed designs and specs through natural collaborative dialogue before any creative work - creating features, building components, adding functionality, or modifying behavior.
• close-session — Use when the user says "close session", "wrap up", "end session", "done for now", "save my work", "commit and push", or when finishing a block of work with no further tasks planned to update project documentation, commit changes, and push safely to GitHub.
• cloud-build-triggers — Use when creating, updating, or managing Google Cloud Build triggers. This skill handles 1st Gen and 2nd Gen GitHub connections, branch patterns, and mandatory IAM validation.
• code-design — Use when starting new feature work, building a new system or module, or when the user asks how something should be designed or architected. Triggers on phrases like 'design this', 'how should we build', 'plan the architecture', 'think through the approach', 'write a design doc', 'what's the right way to implement', or any request that involves deciding how something should work before writing code. Also use when the user provides requirements or a spec and expects a structured design rather than jumping straight to implementation. This skill MUST be used before implementation begins on any non-trivial feature — if the work touches more than one file or introduces a new abstraction, design first.
• dispatching-parallel-agents — Use when facing 2+ independent tasks that can be worked on without shared state or sequential dependencies
• documentation-lookup — Use up-to-date library and framework docs via Context7 MCP instead of training data. Activates for setup questions, API references, code examples, or when the user names a framework (e.g. React, Next.js, Prisma).
• executing-plans — Use when you have a written implementation plan to execute in a separate session with review checkpoints