Documentation / Desktop Ide

Ask vs Agent (and Debug) mode gates

v1.0 Beta

Ask vs Agent (and Debug) mode gates

ACE mutates workspace files in Agent and Debug (Review & Apply only).

Mode	Explains / plans	Applies file edits
Ask	Yes	No
Plan	Yes (plans only)	No
Debug	Yes (investigate + targeted fix)	Yes — Review & Apply only (no auto-apply)
Agent	Yes	Yes (Review and/or auto-apply)

Cursor-style Debug mode

When composer is Debug:

Debug sidebar opens (ready for breakpoints / Start).
Each AI send attaches debug_session context when available: console tail, stack, paused file/line, breakpoints, variables, runtime/target.
ACE diagnoses root cause and may propose minimal file edits.
You confirm with Review & Apply (never silent auto-apply in Debug).
Full project scaffolds stay in Agent.

Ask ACE (Run / Debug)

Exited 0 / clean → Ask, read-only summary.
Run failure → Agent (if plan allows) so ACE can apply a fix.
Debug Ask ACE failure → stays in Debug with live session context + Review & Apply for targeted fixes.

Hard gates

canApplyAiFileMutations() — Agent or Debug (unless Ask ACE success force-read-only).
Auto-apply / lint-fix loops — Agent only.
Plan blocks implementation prompts; Debug only blocks large scaffold requests.

Quick checks

Composer → Debug → Debug sidebar opens; status mentions Review & Apply.
Breakpoint + Debug Start → fail → Ask ACE → mode stays Debug → Review card can appear.
Exited 0 Run → Ask ACE → Ask mode, no disk writes.
Debug mode + “build a full ecommerce app” → CTA to switch to Agent.
Debug mode + “fix this TypeError in greet.js” → diagnose + propose edit → Apply via Review.

Was this helpful?