Agent Skills

Agent Skills let you give TatsuCode reusable, task-specific guidance.

Think of a skill as a focused instruction pack: when to use it, how to approach the task, and what quality bar to enforce.

Agent Skills are separate from AGENTS.md.

AGENTS.md = always-on project context
Skills = optional workflow packs you load when needed

New Here? Start with This

If this is your first time:

Use AGENTS.md for project-wide default behavior (/init creates or updates it from your codebase)
Use Agent Skills for specialized workflows you only need sometimes
Load skills on demand with /skills

A simple mental model:

AGENTS.md = default team rules
Skills = optional specialist playbooks

Why Agent Skills Matter

Without skills, every advanced workflow has to be re-explained in chat.

With skills, you can:

Reuse proven workflows across sessions
Standardize output quality for team tasks
Keep prompts shorter while staying consistent
Scale project-specific conventions without micromanaging every request

Standards and Compatibility

TatsuCode supports industry-standard skill and agent-instruction workflows.

AGENTS.md reference: agents.md
Agent Skills reference: agentskills.io

This makes it easier to share conventions across teams and tools.

Built-In Skills (Current)

Built-In Skill	When to Use
brainstorming	Collaborative design and planning before implementation — gates coding until the design is approved
browser	Browser automation workflows — navigate, click, type, inspect, and screenshot web pages
computer-use	Desktop automation — click, type, drag, scroll any application using native controls
deep-think (experimental)	Complex debugging, architecture tradeoffs, novel problems, high-risk tasks
image-generator	Operational skill for the new image generation pipeline — provider-aware presets, prompt recipes, edit / mask workflows
image-tools	Operational skill for image editing — resize, crop, trim, blur, and other on-the-fly image operations
office-word	Create and edit Word documents (DOCX) with targeted content operations
office-excel	Create and edit Excel spreadsheets (XLSX) with range-level precision
office-powerpoint	Create and edit PowerPoint presentations (PPTX) with slide-level control
skill-editor	Creating, reviewing, and improving skills with clearer structure and trigger criteria
threejs	Three.js / WebGL / WebGPU / React Three Fiber work — backed by 13 deep code-prompt resources covering core architecture, animation, loaders, materials, physics, and post-processing
windows-control	CLI-first Windows automation — PowerShell, COM objects, system utilities, and settings

Cross-Engine Gamedev Resources (shared across Unity and Three.js)

The Three.js and Unity skills share a set of seven cross-engine gamedev resources covering PBR materials, physics, post-processing, color grading, skeletal animation, texture prompting, and visual verification. If you've already loaded one for a Unity task, the same content applies to Three.js work — quality stays consistent across engines.

Quick Guidance

Use brainstorming when planning a new feature or making architectural decisions.
Use computer-use when you need to interact with desktop applications outside the browser.
Use deep-think when a quick answer is likely to miss edge cases.
Use skill-editor when creating/updating skill content and quality rules.

Core Commands

/skills
/skills-refresh

What Each Command Does

/skills — Browse discovered skills and load one
/skills-refresh — Re-scan skills after edits/additions

To author a new skill, ask the assistant directly (e.g. "create a skill for X") — the built-in skill-editor skill will scaffold and write SKILL.md for you. You can also create the folder and SKILL.md by hand under .agents/skills/<name>/ (project) or ~/.tatsu/skills/<name>/ (global), then run /skills-refresh.

Where Skills Live

Primary skill locations:

Project Skills (team/shared)

Store in your repository:

.agents/skills/

Example:

my-repo/
  .agents/
    skills/
      deployment-checklist/
        SKILL.md

Global Skills (personal)

Store in your user profile:

Windows: %UserProfile%\.tatsu\skills\
macOS/Linux: ~/.tatsu/skills/

Use this for personal workflows you want available in every project.

Importing Skills from Websites (Important Safety)

You can import skills from external sources — but treat them like code from the internet.

Safety Rules

Only use skills from trusted websites/authors
Review SKILL.md before loading
Review any bundled scripts or references before use
Do not blindly run commands suggested by unknown skills
Prefer read-only review first, then gradual adoption

Recommended Import Flow

Download/copy skill folder from a trusted source
Place it in project or global skills location
Open and inspect SKILL.md
Validate instructions match your standards
Run /skills-refresh
Load via /skills and test on a low-risk task first

AGENTS.md vs Agent Skills (Separate Systems)

For full AGENTS.md guidance, see AGENTS.md.

Use Case	Better Choice	Example
Team coding conventions	AGENTS.md	naming style, commit policy, test expectations
Specialized release process	Skill	release checklist, staging verification flow
Security review routine	Skill	auth audit and dependency checks
Always-on project architecture rules	AGENTS.md	folder boundaries, layering constraints

Recommended Pattern

Put durable defaults in AGENTS.md
Put specialized playbooks in skills
Load a skill only when needed

Typical Skill Workflow

Ask the assistant to scaffold a skill (the skill-editor system skill handles SKILL.md creation), or hand-create .agents/skills/<name>/SKILL.md
Write purpose + clear “Use this when…” triggers
Keep content concise and practical
Refresh with /skills-refresh
Load from /skills when needed

Practical Prompt Examples

Load deep-think and investigate why this auth flow fails only in production.

Load skill-editor and create a project skill for API contract validation before merging.

Use /skills, load our deployment-checklist skill, and run through release prep.

Skills + Drag & Drop Context

You can combine skills with file/folder drag-and-drop for fast setup:

Drag files/folders into input
Load the relevant skill
Run the workflow with real project context

Great for audits, migrations, and release checks.

Team Best Practices

Keep each skill narrowly scoped
Include clear trigger conditions
Prefer short checklists over long prose
Store team skills in .agents/skills/ under version control
Review skills like any other project artifact

Next Steps

AGENTS.md — project context file guide
Settings — AGENTS.md + project behavior
Commands — full slash command reference
Quickstart — day-one workflow