Modal Overlay Rework

This sprint was the external base for everything that followed. Its brief kept the agent on vendoring discipline: refresh the snapshot, keep the lockfile consistent, carry tests and examples, and only add missing behavior in application-neutral terms. It passed cleanly, so later Gantry-side sprints could rely on the library contract rather than rediscover modal layout primitives.

This sprint translated the refreshed library API into Gantry's own boundary. The task was deliberately not a migration sweep; it was to define the shape later overlays would consume so they stopped owning sizing, chrome, button rendering, and unconditional key swallowing. Its gate stayed green, giving the next sprint a stable application contract to try against real TUI behavior.

This sprint proved the new contract on a real non-text dialog without taking on the full overlay catalog. The brief asked for enough behavior to exercise body layout, scrolling, buttons, chrome, and host key fallthrough. It passed inside the milestone, but the later milestone review showed that proving one call site was not the same as proving every parent overlay dispatch path.

This sprint came first because every later migrated overlay depended on host-level fallthrough being real. The agent did not have to move more surfaces yet; it only had to repair the path from a frontmost shared modal through parent overlay shortcuts into global bindings. It passed cleanly, turning the foundation review's concrete swallowed-help case into a contract later sprints could consume.

This sprint isolated recovery action semantics before modal chrome touched every blocker-class surface. Its scope was intentionally independent of layout: define the order, focus, no-external message, and opaque action tokens once so later prompts could reuse them. The sprint itself went green, but the milestone review later found the order was still wrong in the composed result, which limited how much final confidence this row could carry.

This sprint moved the highest-value prompt family after fallthrough and recovery contracts existed. Its brief covered text prompts, confirmations, fixed choices, and blocker investigation prompts while preserving the engine's prompt-answer wait. The run facts show a replan stage in this area, which fits a migration that had to compose prompt semantics with the newly repaired modal route rather than just repaint a dialog.

This sprint carried stop, halt, error, and local confirmation surfaces after engine prompts had proven the core prompt path. The cut kept the worker focused on stop-class decisions and status bodies: shared sizing, scrollable reports, recovery ordering, and preserved local confirmation behavior. It stayed green, so the read-only migrations could proceed without reopening stop semantics.

This sprint grouped the smaller read-only dialogs so a fresh agent could remove a cluster of bespoke panel math without also solving pager boundaries. Help, info, and session-style content all needed shared overflow behavior and the repaired global-key path. The row stayed green, which left the richer pager audit as a separate problem.

This sprint handled the ambiguous read-only surfaces after simpler dialogs had moved. Its brief asked the agent to migrate modal-shaped pager presentations while recording which overlays were genuinely whole-body panes. That made the boundary review explicit: stats and git activity could stay panes, while their dialog-shaped drill and pager targets had to obey the shared modal contract.

This cleanup sprint came last because deleting old modal behavior is only safe once every migrated family has a shared replacement. Its gate ran more than once but stayed green, which marks the cleanup as broad rather than architecturally blocked. The later milestone review still caught recovery ordering outside this cleanup's declared scope, so this row closes legacy rendering paths rather than the entire plan.

This sprint was the run's only recorded recovery. The first attempt did not settle the blocker-class order strongly enough, review noted it, and Gantry replanned this exact sprint before the milestone continued. That makes the boundary clear: modal history could not build on blocker, halt, and error prompts until their producers, buttons, focus, and tests all agreed on one action contract.

This sprint created the data contract for reopening without touching the rendering workflow yet. The brief kept the work bounded to Recent events: store display-ready modal content, prompt identity, liveness, truncation state, and current choice tokens while preserving the existing bounded feed. It passed cleanly because the reopen decision and UI flow were left for the next slice.

This sprint turned the stored payload into the user workflow. It could stay focused because the previous sprint had already defined the snapshot contract and the first sprint had repaired recovery choices. Its job was to route selection from Recent events either back to a live prompt or into a read-only modal, while preserving feed selection for entries that are not modal content.

This final sprint was verification, not a new modal system. It consumed the repaired recovery contract and the reopenable payload contract, then checked the original complaints across live and reopened paths. It stayed green, closing the run with focused regression coverage for clipped text, swallowed global keys, and wrong blocker recovery ordering.