CLI Alias Updates

This first sprint was intentionally narrow: add print-mode spellings to the existing headless boolean rather than create a second execution mode. That made it a good opener because every downstream caller already knew how to consume headless mode. The gate and review stayed green, which fits a compatibility alias whose main risk was flag collision and routing drift. Later resume work could reuse the same short-flag collision testing posture.

This sprint translated flag-first resume muscle memory into Gantry's existing bare resume verb. Its boundary was early argv normalization: after the alias was stripped, normal build flags and engine dispatch could stay untouched. The run facts show a clean pass, so named resume aliasing did not force a repair or a broader command refactor. No-name resume was left to the next sprint, keeping this worker scoped to explicit targets.

This sprint put the decision core for unnamed resume ahead of any interactive UI. The brief separated interactivity and most-recent intent into explicit inputs, so a fresh worker could test the TTY matrix without needing a terminal. The recorded gate ran cleanly more than once for this unit, which points to extra verification rather than a failed repair loop. It established the scriptable behavior that the picker had to respect later.

This sprint added the TTY-only picker after deterministic selection had already defined when a picker was allowed. That cut kept script safety out of the UI layer: the picker appears only after the resolver has decided interaction is appropriate. The unit stayed green in the run record, so the row formatting, selection plumbing, and cancel behavior fit inside one sprint. Its milestone review later clarified that this picker remained resumable-run scoped, not a repo-wide history browser.

This sprint closed the CLI milestone by making the new spellings discoverable while keeping canonical commands primary. It was a documentation and usage-output pass, placed after behavior so help could describe what the binary actually did. The run stayed green, but the milestone review still found a scope boundary around recent-run breadth. That made the docs handoff important: later TUI work had to share terms without overstating the CLI picker.

This sprint removed a naming collision before adding a real recent-runs affordance. Its brief was low-risk and behavior-preserving: rename the misleading Agents menu wording so it described current-run stage logs. The run facts show a clean pass apart from an extra green gate, so the change was verified without repair. That gave the following recent-job work clearer terminology to build around.

This was the milestone's stress point. The sprint had to define a read-only recent-job presentation model that could include historical runs without broadening the CLI resume resolver. Its first gate went red and a fix pass ran before the unit passed, which locates the hard part in getting filtering, ordering, bounds, row tags, and empty state correct. The repair kept the model distinction explicit instead of collapsing back onto the resumable-only picker.

This sprint wired the recent-job rows into the Run menu as an inspection affordance, not a live monitor switch. The brief held the boundary tightly: selecting another run should open read-only status and offer a copyable resume command, while the current monitored run stayed unchanged. The recorded unit stayed green, suggesting the model from the prior repaired sprint was stable enough for UI wiring. It left engine hot-swap explicitly out of scope.

This final sprint reconciled docs and fixture-backed menu references after the TUI changes. Its job was to explain the difference between the read-only recent-job menu and the CLI picker, refresh menu wording, and avoid claiming that CLI resume listed every historical run. The run stayed green, so the closeout was not another implementation repair. It made the milestone's distinction explicit before the top-level merge.