refactor(tui): prompt deep-module PR 1 — frame snapshot + height budgets (#218)

* chore(tasks): record delegation-lane and verification lessons

* docs: prohibit deferring applicable fixes

* test(tui): characterize prompt compatibility contract

* docs(tui): design content rendering standardization

* docs(tui): plan content rendering standardization

* refactor(tui): snapshot prompt delegates once per frame

* fix(tui): keep prompt scenes within terminal height

Add prompting/renderer.py with PromptSceneBudget and a priority-ordered
row allocator (modal > input > footer > pinned > body > status >
shortcuts). Rendering is unchanged for scenes that fit; on overflow the
scene is clipped tail-first per priority, pinned allocation is
zero-safe, and the interactive preview no longer derives a competing
body budget.

* test(tui): cover mid-turn terminal shrink in PTY layout e2e

Resize the live PTY through heights 12/8/6/4 during a running turn and
assert the redrawn frame stays within the new geometry, never
fossilizes the input card, and the turn still completes after the size
is restored.

* chore(tasks): record codex-lane operating contract lessons

* fix(tui): refresh shell-mode footer budget and strengthen resize assertion

Shell-mode rendering never set _prompt_footer_row_budget, so
_fit_toolbar_to_terminal clipped the bottom toolbar against a stale
agent-mode value (or the initial 0, hiding the footer entirely in a
pure shell session). Refresh it every shell render. Replace a
trivially-true pyte row-count assertion in the resize e2e with a
width-overflow check.

* fix(tui): refresh shell-mode update-notice frame snapshot

The per-frame _prompt_frame_update_notice snapshot was refreshed only by
the agent render path and never cleared with the frame, so shell frames
suppressed a live update notice (initial None) or replayed a stale
agent-mode notice after a mode switch. Refresh the snapshot in the shell
render like the agent path and reset it in _clear_prompt_frame.

* test(tui): cover after-render update-notice snapshot cleanup

Extract the after_render frame-clear closure into a named
_clear_prompt_frame_snapshot method so the lifecycle cleanup is directly
testable, and assert a completed frame leaves _prompt_frame_update_notice
as None.

Author: elkaix

AGENTS.md

@@ -41,6 +41,11 @@ contract and C01–C15 tripwires below; ship the matching tests and verification
   `uv run ...` or `uv run --directory <package> ...`.
 - **Keep changes surgical.** Do not perform drive-by refactors, formatting churn, dependency
   upgrades, or generated-file rewrites unless the task requires them.
+- **Never propose deferring an applicable issue or fix.** If a problem is real and a fix applies,
+  do the fix now — even when it requires a redesign or significantly more work. "Defer",
+  "follow-up later", "out of scope for now", and equivalents are prohibited recommendations; the
+  only permitted alternative to fixing immediately is stopping to request expanded scope, then
+  fixing.
 - **Do not expose secrets or PII.** Never print, commit, or copy API keys, OAuth tokens, session
   data, user config, or logs that may contain credentials.
 - **Do not add new telemetry, hosted endpoints, external services, or third-party dependencies**

CHANGELOG.md

@@ -15,6 +15,7 @@ GitHub Releases page; `0.8.0` is the new starting line.
 
 ## Unreleased
 
+- Freeze the public shell prompt compatibility contract with constructor and rendering coverage.
 - Add xAI Grok OAuth login (browser loopback and device-code).
 - Add GitHub Copilot device-code OAuth login for individual github.com accounts.
 - Add DigitalOcean Gradient AI browser OAuth login with dynamically discovered Inference Routers; router-discovery failures (unauthorized, outage, malformed, empty) are now reported distinctly instead of silently yielding no models.
@@ -24,6 +25,7 @@ GitHub Releases page; `0.8.0` is the new starting line.
 - Fix queued follow-up input showing a bordered ghost; pressing Enter during an active turn now shows one intentional queued row.
 - Harden provider OAuth credential handling: scope token refresh to the active/selected provider, serialize persistence under a lock with atomic config replacement, roll back replaced credentials when a save fails, fail closed on credential migration, and validate OAuth token and implicit-state responses.
 - Fix an intermittent doubled/"ghost" copy of the running-prompt block (agent tree, spinner, and tip) on long streaming turns: after a scrollback handoff the suppressed live body now waits for the terminal's re-requested absolute cursor position to settle before it re-expands, so it repaints against a correct cursor model instead of the mis-anchored frame left behind by `run_in_terminal`.
+- Fix the "update available" footer notice being suppressed on shell-mode frames (and a stale agent-mode notice replaying after a mode switch): the shell prompt render now refreshes the per-frame update-notice snapshot like the agent path, and the snapshot is cleared with the frame.
 
 ## 0.60.0 (2026-07-18)