Context
The Slant4D Code Mode path should follow the Cloudflare Code Mode API shape exactly: one search API and one execute API. Agents discover the current tool names/schemas through search, then write JavaScript for execute. The Slant4D product/runtime side is already being handled separately; this issue tracks keeping the public skill/plugin guidance aligned with that contract.
Skill requirements
- Describe Code Mode as a two-API JavaScript surface: search and execute.
- Avoid implying that every Slant4D operator tool is exposed as a top-level MCP tool in Code Mode.
- Keep client-specific wrapper names such as portal_codemode_search and portal_codemode_execute clearly labeled as harness exposure, not the underlying product API concept.
- Keep direct operator MCP guidance separate from Code Mode portal guidance.
- Verify against live MCP metadata before publishing skill updates.
Current observed ergonomics
- Search returns the current JavaScript-callable upstream tool names and schemas.
- Execute runs JavaScript against those discovered tools.
- Responses may be nested or text-wrapped JSON, so agents should unwrap/parse before making decisions.
- Broad record dumps and batched proof calls can truncate or timeout; skill guidance should prefer bounded execute scripts.
Context
The Slant4D Code Mode path should follow the Cloudflare Code Mode API shape exactly: one search API and one execute API. Agents discover the current tool names/schemas through search, then write JavaScript for execute. The Slant4D product/runtime side is already being handled separately; this issue tracks keeping the public skill/plugin guidance aligned with that contract.
Skill requirements
Current observed ergonomics