Summary

Built-in /claude-api does not appear to receive or act on the prompt text that follows the slash command, unlike other built-in skills.

Problem

When invoking the built-in Claude API skill with inline prompt text, Warp appears to drop or ignore the remainder of the prompt instead of passing it through to the skill. In the reported example, the user entered /claude-api Let's build a sentry monitor with the claude sdk, but the response only acknowledged previously provided rules/skill constraints rather than addressing the Claude SDK request.

Expected behavior

The built-in /claude-api skill should receive the text after the slash command and respond to that request, consistent with how other built-in skills handle inline prompt text.

Actual behavior

The response does not address the inline Claude SDK request. It appears to behave as if the trailing prompt text was not forwarded to the skill.

Reproduction steps or desired workflow

1. Open Warp Agent mode.
2. Invoke the built-in Claude API skill with an inline request, for example: /claude-api Let's build a sentry monitor with the claude sdk.
3. Compare the result with other built-in skills that do act on the remainder of the prompt.
4. Observe that /claude-api does not address the inline request and instead responds only to surrounding context.

Artifacts

Screenshot described in user report but not attached in the filed issue. The screenshot showed /oz-platform how does scheduling work? producing a relevant answer while /claude-api Let's build a sentry monitor with the claude sdk responded only with a generic acknowledgement about rules and listed skills.

Warp version

v0.2026.04.27.15.32.stable_03

Operating system

macOS 26.3.1