Permission prompts still appear with --dangerously-skip-permissions + skipDangerousModePermissionPrompt: true

Summary

Permission prompts continue to appear in the desktop/IDE UI for tool calls despite both:

• The CLI being launched with --dangerously-skip-permissions
• "skipDangerousModePermissionPrompt": true set in ~/.claude/settings.json

Expected: no prompts at all — that is the user-visible contract of the flag name.

Actual: certain tool calls (observed with Bash for things like git push --force-with-lease, gh run rerun, and running background gh run watch tasks) still trigger a UI prompt, breaking auto-mode / long-running flows.

Impact

• Interrupts batch work the user has explicitly opted into.
• Erodes trust in the flag — user has explicitly chosen to accept risk.
• Makes automation (scheduled tasks, /loop dynamic mode, chained tool calls) unreliable.

Repro hints (approximate)

• Launch claude --dangerously-skip-permissions with settings.json containing "skipDangerousModePermissionPrompt": true.
• From within a normal session, have the agent issue a Bash tool call like gh run rerun <id> or git push --force-with-lease ....
• A prompt still appears in the UI.

Environment

• Claude Code 2.1.111 (via desktop-bundled claude-code)
• Platform: darwin / macOS
• Permission mode: auto with --allow-dangerously-skip-permissions

Ask

Respect the flag uniformly — if the user has opted into dangerous skip, no tool-use category (Bash, git operations, background process launches) should re-prompt in the UI layer.