Survey Phase 2, and fix what "done" means

[Seungpyo1007]

Track B of the 0.3.0 plan, first deliverable. Survey only — no code changes.

The finding that matters more than the table

Two documents define "done" for Phase 2, they disagree, and neither matches the code:

• DESIGN.md — Phase 2 is "fork Code-OSS", because visual effects were thought impossible through the extension API.
• README — Phase 2 is "deepen renderer-level visual effects at the editor core", marked in progress.
• Reality — this is not a fork. It is Electron 43 + React 18 + Monaco with its own renderer from the start. The reason the fork existed — no access to the renderer — never applied.

So Phase 2's original goal was met by another route. What remains is not the fork question but whether the four pillars in DESIGN.md §2 are actually on screen.

The table

Verdicts are from reading the code. △ means something similar exists but not where the design put it.

Pillar	Verdict
2.3 Agent activity panel	Done. Current action, plan view, tool log, token meter — all four filled by real paths.
2.1 Editing animation	Glow and afterglow ✓, single undo step ✓, ghost cursor missing, streaming insertion △
2.2 Diff visualization	Gutter colours △ (they track git, not the pending proposal), rationale △ (on the card only), CodeLens missing — zero registerCodeLensProvider calls
2.4 Multi-file overview	Badges, progress dots and jump ✓; a bar list rather than the card grid △

Definition of done

Three things the design named that the code does not have:

1. Ghost cursor — small, decoration-based, immediately visible.
2. Rationale tooltip — rationale already exists; it needs to hang on the changed line.
3. CodeLens accept/reject — hunk selection logic already exists; this puts the buttons in the editor.

All three sit on the decoration layer built in #70.

The remaining △ items are places where the app already answers the same need a different way (full diff in its own tab, file overview as a list). Reshaping those to match the prose would be a separate UI decision, not Phase 2 completion. The fork item closes as not-applicable.

"Streaming insertion" is the one genuinely open question: today the animation replays already-complete text, so it has no relation to how fast the model actually writes. Streaming real tokens would mean changing the proposal/approval flow, not the visuals. Left as separate work.