The Layout
- Left: the skill’s files.
SKILL.mdholds the skill’s definition and instructions; procedural skills also havemain.py(or a Node.js entry point),requirements.txt, and any supporting files. - Center: the editor, with syntax highlighting for Python, Markdown, JSON, and YAML.
- Right: tabbed panels for Validation, Test Runner, Credentials, Metadata, and Versions.
Validation
The Validation panel checks the skill continuously as you edit. Checks include a valid frontmatter schema, a present entry point, declared dependencies, safe operations only, and that the skill’s declared inputs and outputs match what the code actually reads and writes. A failing check blocks Publish until fixed, so you cannot ship a broken skill by accident.Test Runner
The Test Runner executes the skill in a sandbox without publishing:- Provide sample input as JSON, or leave it empty for a connectivity check
- Click Run
- Output streams in real time, with a status badge (Running, Success, Failed) and elapsed time
Refine with AI
The Refine input at the bottom of the editor asks AI to edit the active file for you: “Add a totals row to the report table” or “Handle the case where the API returns no results”. The AI edits in place; you review the change in the editor.Credentials and Metadata
- Credentials tab: attach a stored credential to the skill and test it.
- Metadata tab: edit the skill’s name, description, icon, category, scope, run budgets, and any tunables the skill exposes.
Platform tools and connected services
The Runtime panel lists the platform tools the skill declares. A skill gets extra abilities through theplatform_tools list in its frontmatter:
- Built-in tools such as
memory_search(search your organization’s Files),web_search, andcreate_file. - Capabilities such as
gmail.readorcalendar.read. - Tools from Connected Services. Any tool from a connected MCP service works here too. Add its name (for example
mcp_rankpilot_search) toplatform_toolsand the skill can call that service during its run. The exact names are listed on the Connected Services screen when you expand a service’s row. Connector calls from a skill run with the same organization isolation and audit trail as calls from chat.