A Claude Code Skill Collection for Turning Reusable Context Into a Second Brain

The problem is not capability, it is reusable context

Knowledge work often breaks down at the handoff between a general-purpose assistant and the particulars of a team's work. A model may be able to draft a runbook, shape a presentation, or call an external tool, but it still needs the operating procedure, brand rules, integration details, and known pitfalls for that specific task.

MuLIAICHI/second-brain-skills-ml is a collection of custom Claude Code skills aimed at that gap. The repository frames the collection as a way to turn Claude Code into a “second brain” for knowledge work. Its practical claim is more modest and useful: organized instructions and resources can make an agent more effective at recurring, specialized workflows.

The project is likely most relevant to people already using Claude Code who want repeatable help beyond software implementation. It groups workflows for external integrations, documentation, presentations, programmatic video, brand definition, and the creation of additional skills.

A collection organized around task-specific expertise

The repository's key design idea is progressive disclosure. In plain language, that means keeping detailed task guidance out of the active context until it is relevant. Rather than loading every instruction and tool definition for every request, Claude loads the material associated with the skill being used.

That matters especially for integrations. The MCP Client skill is designed to connect Claude Code to Model Context Protocol servers, including services such as Zapier, GitHub, and Sequential Thinking. MCP is a protocol for exposing tools to an AI client. Tool schemas can be large, so the skill's documented approach is to load the schemas on demand instead of keeping them in context continuously.

The setup described in the README uses an MCP configuration file with server definitions. It distinguishes remote servers authenticated with a URL and API key, local standard-input/output servers launched by a command and arguments, plus SSE and streamable HTTP endpoints. The skill also documents a simple operational habit: test newly connected tools and record quirks in a project's CLAUDE.md file.

This is a practical lesson for any agent integration. A schema can describe the shape of a call, but it may not capture every constraint that appears in real usage. Keeping project-local notes about required formats, rate limits, or surprising behavior gives later interactions concrete context without requiring the same discovery work each time.

Documentation as an executable operating pattern

The SOP Creator skill focuses on runbooks, playbooks, technical documentation, checklists, decision trees, handoff documents, and onboarding material. Its stated premise is that long documents are less useful than guidance people can scan and act on.

Its proposed structure puts a Definition of Done near the top, followed by when to use the document, prerequisites, numbered process steps, completion checks, failure handling, and a place for questions. The writing rules are similarly operational: make steps action-first, state numbers and thresholds rather than vague frequencies, put warnings before risky actions, and make decisions explicit with conditions.

There is a useful connection here between documentation and agent skills. Both are forms of operational knowledge. The repository treats that knowledge as something that can be packaged for a person to follow or for Claude Code to apply when the relevant request arrives. The value is not that the agent has a generic ability to write prose. It is that a recurring process gets a stable, task-shaped structure.

Brand configuration feeds presentation work

A separate Brand & Voice Generator skill creates a set of files intended to become a source of truth for content and presentation work. The README lists brand.json for colors, fonts, and assets; config.json for output settings; brand-system.md for design guidance; and tone-of-voice.md for writing style.

Those artifacts feed into the PPTX Generator skill, which the repository describes as a tool for on-brand presentations and LinkedIn carousels. It documents three modes: conventional 16:9 slide generation, square carousel generation that exports to PDF, and layout management for a layout cookbook.

The presentation workflow emphasizes choosing a visual layout before defaulting to a text-heavy content slide. For example, equal items map to a multi-card layout, a small set of large figures to a statistics layout, comparisons to two columns, and a central idea with surrounding elements to a circular composition. The repository also documents batch generation of up to five slides at a time for reliability and rules intended to avoid repetitive layouts.

The implementation lesson is clear: separating brand data from generation instructions makes a reusable asset out of design decisions. A team can define its palette, typography, visual principles, and voice once, then apply them across outputs. The tradeoff is setup. The README explicitly marks the presentation workflow as requiring a more complete brand configuration than the SOP or Skill Creator workflows.

Video and skill-building extend the same model

The Remotion Video Creator skill covers programmatic video using React and Remotion. The repository describes videos as React components and points users toward a separate Remotion project for previewing and exporting. Its documented areas include animation, compositions, assets, captions, charts, 3D, transitions, and maps.

The guidance also narrows the unit of work: build one composition at a time rather than requesting an entire video in one step. That is a sensible constraint for iterative visual work, where preview and revision are part of the process.

Finally, the Skill Creator turns the collection's organizing principle back on itself. It presents skills as modular packages with a required SKILL.md and optional scripts, references, and assets. Its advice is to add only the context Claude does not already have, choose the right level of specificity for a task, and use progressive disclosure across metadata, instructions, and supporting resources.

That makes this repository less like a single application and more like a catalog of working patterns. Some skills can be used immediately, according to the README, while others need configuration, external projects, or integration credentials. The common thread is deliberate context design: document the reusable details, scope them to a task, and make them available when the task actually appears.

For builders exploring that approach in Claude Code, inspect the repository and its documented setup paths at MuLIAICHI/second-brain-skills-ml.