Skill workflows stall or run away in the app across Cursor and Anthropic models

[devswha]

Summary

CUA testing of every bundled skill exposed workflow execution failures in the web app. Skill selection and the Run confirmation gate work, but the executable workflows do not reliably reach their terminal user-facing state.

Environment

• Gajae Code App v2.0.0-beta.3
• Disposable git project with only README.md
• Models tested: cursor/cursor-grok-4.6-xhigh-fast, anthropic/claude-fable-5
• App dev server on macOS arm64

Results

Skill	Model	Result
deep-interview	cursor-grok	Failed to ask the requested first question after more than 60 seconds; repeatedly probed gjc deep-interview read, which gjc, shell redirects, and CLI help while reporting missing/incomplete command output. Had to stop manually.
ralplan --deliberate	cursor-grok	Received the full objective, but never produced a receipt or approval gate. Repeatedly probed pwd, ls, which gjc, environment variables, and find, while reporting that normal shell output was empty. Had to stop manually.
ultragoal	claude-fable-5	Created and byte-verified the requested one-line file, but first invoked the goal tool as {} and hit validation failure (op missing). It then continued through an oversized quality-gate flow, displayed a quality-gate write error, consumed about 49,290 tokens, and was still running after the requested file was verified. Had to stop manually.
team	n/a	Expected behavior passes: it is hidden from the picker and typed /skill:team is refused with the tmux/ultragoal guidance.

Reproduction

1. Add a minimal local git project.
2. Start a new task.
3. Pick one of the visible bundled skills from the sparkle menu.
4. Add a bounded objective after the inserted /skill:<name> command.
5. Click Send, then Run.
6. Observe that deep-interview/ralplan repeatedly diagnose missing CLI or shell output, or that ultragoal continues far past verified completion.

Ultragoal goal-tool error:

Validation failed for tool "goal":
op: Invalid option: expected one of "create"|"get"|"complete"|"resume"|"drop"|"pause"
Received arguments: {}

Expected

• deep-interview reaches one user-facing question without repeatedly probing its own CLI.
• ralplan produces its persisted receipt and approval boundary.
• ultragoal calls goal with valid arguments, completes once the bounded goal and quality checks pass, and does not consume tens of thousands of tokens for a one-line file.
• Provider tool results and workflow state are delivered consistently enough that the model does not self-diagnose missing shell output.

[devswha]

App-side command transport and CI blockers have now been remediated in #7–#9. The remaining deep-interview, ralplan, and ultragoal failures are owned by the pinned coding-agent workflow/tool lifecycle. No app-side fallback executor was added. Remaining work requires direct SDK reproduction and an upstream fixed release, followed by the acceptance scenarios already listed in this issue.