Open Settings → Agent → Skills & Context files.
Enable or disable skills per workspace so each agent follows the right project-specific instructions.
WORKSPACE.md
stagewise auto-generates a WORKSPACE.md file in .stagewise/ for each workspace. It contains structured project analysis and is kept up to date across sessions.
Generate or update WORKSPACE.md
- Open Settings → Agent → Skills & Context files
- Select the workspace
- Click Generate (or Regenerate to update)
The file is written to .stagewise/WORKSPACE.md in your project root. Commit it to source control.
AGENTS.md
An optional AGENTS.md file in your project root provides custom instructions — coding conventions, preferred libraries, project terminology.
AGENTS.md is opt-in and disabled by default. It can get outdated and mislead the agent. Use skills instead for project-specific instructions.
Skills
Skills are SKILL.md files that extend the agent with structured instructions for specific tasks. Learn the concepts →
Skill locations
Skills load from two places in priority order:
| Location | Priority |
|---|
.stagewise/skills/ | Highest |
.agents/skills/ | Lower |
Enable or disable skills
In Settings → Agent → Skills & Context files, toggle individual skills on or off per workspace. Disabled skills are ignored even if the files exist on disk.
Create a skill
Create a folder with a SKILL.md file:
---
name: "my-skill"
description: "When and how the agent should use this skill."
---
# Instructions
Your structured instructions here...
description field tells the agent when to activate the skill. Be specific:
---
name: "tailwind-patterns"
description: "Use when creating or modifying React components styled with Tailwind CSS."
---
# Instructions
Always use design tokens from `theme.ts`, never hardcoded colors.
Components should accept a `className` prop and merge it with defaults.
Skill folder structure
.stagewise/skills/my-skill/
├── SKILL.md # Instructions (required)
├── references/ # Files the agent can read
└── assets/ # Supporting files
scripts/ folder are read-only — the agent can reference but not execute them.
