Port upstream 0.65.0: replay Claude CLI cursor redraws onto a virtual screen

[Finesssee]

Summary

Claude CLI /usage panels redraw differentially: the CLI moves the cursor and rewrites only changed cells. The local parse_cli_output used strip_ansi, which drops every cursor jump, so the fixture frame yields 51%usd and the model-specific weekly row is lost. The capture is now replayed onto a 160x50 virtual screen (the same geometry as the PTY in tty_runner.rs) before parsing, so unchanged cells of earlier frames are kept and erased content is never reused. Plain and color-only reports come back unchanged (preservePlainReports mode), so CR-delimited legacy output is not clipped or wrapped.

The replay is used in parse_cli_output, in the workspace-trust-prompt check, and for the CLI error-marker check in fetch_via_cli (rendered once, then passed on; rendering is idempotent on rendered text).

Upstream reference

• Release note: "Claude: preserve model-specific weekly quotas and account identity when CLI panels redraw with cursor jumps" (fix(claude): replay CLI PTY output onto a screen before parsing (#3746) steipete/CodexBar#3822, fixes Fable doesn't show reliably. It's really unpredictable. No way to know if it'll show or not. steipete/CodexBar#3746; 53d46c836 / fix(claude): replay CLI frames before parsing usage steipete/CodexBar#3854), v0.65.0.
• Tag-pinned sources: Sources/CodexBarCore/Providers/Claude/ClaudeCLIScreen.swift, ClaudeStatusProbe.swift (cleanCapture), Tests/CodexBarTests/ClaudeCLIScreenTests.swift, ClaudeCLIScreenProbeTests.swift, Tests/CodexBarTests/Fixtures/Providers/Claude/{usage,status}-pty-differential-redraw.ansi.

Ported / Deferred

Ported:

• New rust/src/providers/claude/cli_screen.rs (697 lines, no new dependencies): tokenizer equivalent to upstream's ESC(?:\[...|\]...|.)|[^ESC]+ pattern (OSC consumed whole, BEL/ST/end terminators), CSI A B C D G H f K J with parameters capped to two and clamped, K/J modes 0-2 only, CR/LF/CRLF/BS/tab, wide-glyph cells with trail cells, combining marks attached to the previous cell, wrap at column 160, scroll after row 50.
• Wide-glyph width: the crate has no width/grapheme dependency (unicode-segmentation is only transitive), so upstream's width rule is ported directly (wide ranges, emoji presentation, U+FE0F emoji, zero-width marks). Grapheme clustering is a small approximation of UAX fix: resolve clippy warnings #29 (marks, ZWJ sequences, emoji modifiers, regional-indicator pairs, CRLF) rather than full segmentation.
• Fixtures: usage-pty-differential-redraw.ansi and status-pty-differential-redraw.ansi, byte-exact from v0.65.0 (verified with cmp), under rust/src/providers/fixtures/claude/. .gitattributes marks them -text so the CR bytes survive EOL conversion.
• All vectors from ClaudeCLIScreenTests.swift, plus the plain-report cases from ClaudeCLIScreenProbeTests.swift.
• Local adjustment required by the fixture: current /usage panels print a "Session" summary (Total duration, Usage: ... cache read) above the plan limits. is_cli_activity_stats_response matched it and rejected the whole capture. It now yields to a current session / current week section (has_plan_limit_section); stats-only captures are still rejected (existing tests unchanged).

Deferred / not applicable:

• /status identity probe and ClaudeCLISession Escape-before-reuse (Claude CLI source: /status identity probe never runs because the reused session still has the /usage panel open (email/plan badge stay empty) steipete/CodexBar#3821): Windows has no /status probe and spawns a fresh PTY per fetch. The status fixture is used only to check that a redrawn identity block renders to the final frame and that the existing email/login-method extraction reads it.
• Scoped window title: upstream reports "Fable only"; the local scoped-weekly parser keeps the panel label verbatim, so the title is "Fable" (id claude-weekly-scoped-fable, 51% used match). Left as is (pre-existing local behavior, not part of this item).

Validation

Run with the pinned toolchain and a process-local CARGO_TARGET_DIR (the pwsh wrapper was refused in this worktree):

• cargo +1.98.0 fmt --all clean.
• cargo +1.98.0 clippy --workspace --all-targets -- -D warnings: pass, no warnings.
• cargo +1.98.0 test -p codexbar providers::claude: 180 passed, 0 failed, 1 ignored (pre-existing native-Claude test).
• cargo +1.98.0 test -p codexbar (full): 2176 passed, 0 failed, 1 ignored; bin test 1 passed.
• New tests (claude::cli_screen::tests, 16): 33 upstream vectors, joiner/flag width, OSC edge cases, synthetic differential frame vs golden, backspace, clear-screen no-resurrect, overflow parameters bounded to 50x160, wrap/scroll, range-table sortedness, plain-report preservation (CR delimiters, OSC contents, no clipping past 50 rows or 160 columns), fixture: session 97% left, weekly 73% left, reset "Resets Sep 23 at 3pm (Europe/Stockholm)", scoped Fable 51% used, and strip_ansi alone yielding 51%usd.

Affected areas

• Rust backend (rust/src/providers/claude)
• Tauri shell
• React UI / tray / settings / float bar
• Fixtures / .gitattributes
• CLI

UI proof

Not applicable.

[coderabbitai]

Important

Review skipped

Auto reviews are disabled on base/target branches other than the default branch.

Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Advanced

Run ID: 66f96b9b-6d97-46c1-9009-0bf954f0f45f

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[Finesssee]

Adversarial validation (lane-A) at 200f610

Review verdict: the branch at its published head matches the 0.65.0 audit PR-2 spec; local HEAD equals the remote head (200f6107), so no re-merge or fix commit was needed. The replay (cli_screen.rs, 160×50 virtual screen, CSI A/B/C/D/G/H/f/K/J + CR/BS/tab/OSC, wide-glyph/combining handling, preserve_plain_reports) is wired before Claude CLI parsing, so cursor-positioned differential redraws keep unchanged cells; tests include the upstream synthetic vectors and the two differential-redraw ANSI fixtures.

Checks at 200f610 (CARGO_TARGET_DIR=W:\cargo-target\lane-a, jobs=4, RUST_TEST_THREADS=4):

• cargo fmt --all --check: pass.
• cargo clippy both manifests --all-targets -- -D warnings: only the 3 documented pre-existing main-drift findings; 0 in this PR's diff.
• cargo test rust manifest: 2176 passed / 0 failed / 1 ignored (claude focused: 223/0).
• cargo test desktop manifest: only the documented Isolate bootstrap payload test from real settings #684 bootstrap_payload_exposes_every_provider_variant failure (env-dependent baseline on branches without Make the bootstrap catalog test hermetic (#684) #711; expected, hermetic fix lives on release/v0.70.0).

No new code was required; validation-only item.