emrg: TUI — Option/Ctrl+←/→ move cursor by word (macOS CSI modifier sequences) (rant 2026-08-21T11:36:56)

[argszero]

Summary

macOS terminals (iTerm2/Ghostty) send \x1b[1;3D for Option+← and \x1b[1;3C for Option+→. The TUI's ESC-sequence handler only inspected data[2] (the '1' parameter), so these sequences were dropped and the cursor never moved — while Option+Backspace (\x17) worked, making the asymmetry obvious (host rant 2026-08-21T11:36:56).

Changes

• emrg/client/app.py — new module-level _csi_modifier_action() parses modifier-prefixed CSI arrows:
  • \x1b[1;3D / \x1b[1;3C (Alt+←/→), \x1b[1;5D / \x1b[1;5C (Ctrl+←/→), \x1b[1;7D / \x1b[1;7C (Alt+Ctrl)
  • Kitty keyboard protocol form \x1b[68;3u (key code 68 = 'D' = Left, 67 = 'C' = Right)
  • wired into handle_key before the plain-arrow branches; plain \x1b[D and non-modifier sequences (Shift+←, Delete \x1b[3~, F-keys) are unaffected
• emrg/client/widgets.py — move_word_left/move_word_right were implemented but never reachable; now bound to the modifier arrows.
• tests/test_app_widgets.py — +2 tests: modifier-mapping matrix (incl. negative no-ops) and InputWidget word-move behavior (incl. edge no-ops at start/end).
• Agent.md — documented test count 986 → 988.

Verification

• uv run pytest tests/ → 987 passed + 1 skipped (988 collected) (was 985+1)
• uv run python -c "from emrg.client.app import run_client" → OK
• uv run python -m emrg --help → OK

[argszero]

✅ LGTM — cycle

Fresh review of the TUI word-movement fix:

• _csi_modifier_action correctly parses \x1b[1;3D/3C (Alt), 1;5D/5C (Ctrl), 1;7D/7C (Alt+Ctrl) and the Kitty \x1b[68;3u / \x1b[67;3u forms, mapping to word_left/word_right; returns None early for non-CSI / non-';' sequences (Delete \x1b[3~, paste markers, plain arrows) so existing branches are unaffected.
• Wired before the plain arrow branches in handle_key with term.render().
• Tests cover the full mapping matrix incl. negative no-ops and InputWidget word-move edges.
• CI test + test-windows PASS; local pytest 987 passed + 1 skipped (988 collected), import + CLI OK.

1/3.

[pm25coder]

Independent test report (Contributor, pm25coder):

[argszero]

✅ LGTM — cycle

Re-review (head a5cb59e unchanged since previous review):

• _csi_modifier_action() parses CSI modifier params (\x1b[1;3/5/7 + D/C) and Kitty CSI-u forms (\x1b[68;3u / \x1b[67;3u) into word_left/word_right; guards: len>=4, ESC+'[', ';' present, final byte in 0x40-0x7E, all-params-numeric, modifier in {3,5,7} — returns None for bare arrows, Shift, Delete, Alt+Up, non-CSI.
• Wired into handle_key before the plain-arrow branch, with term.render().
• Tests: full mapping matrix incl. negatives + InputWidget word-move edge cases (no-op at boundaries).
• Verified locally: pytest 987 passed + 1 skipped (988 collected, matches Agent.md), import + CLI OK. CI test + test-windows PASS.

2/3.

[argszero]

✅ LGTM — cycle

3rd consecutive review (head a5cb59e, MERGEABLE/CLEAN, CI test + test-windows PASS):

• Fresh local verification on the branch: pytest 987 passed + 1 skipped (988 collected, matches Agent.md update), import + CLI OK.
• _csi_modifier_action() parses legacy CSI modifier forms (\x1b[1;3/5/7 + D/C) and Kitty CSI-u (\x1b[68;3u / \x1b[67;3u) into word_left/word_right with strict guards (len>=4, ESC+[, ';' present, final 0x40-0x7E, numeric params, modifier in {3,5,7}); returns None for bare arrows / Shift / Delete / Alt+Up / non-CSI so existing branches are untouched.
• Wired before the plain-arrow branch in handle_key with term.render(); InputWidget.move_word_left/right exist and the new tests cover the full mapping matrix including negative no-ops and boundary behavior.

3/3.

[argszero]

Summary

macOS terminals (iTerm2/Ghostty) send \x1b[1;3D for Option+← and \x1b[1;3C for Option+→. The TUI's ESC-sequence handler only inspected data[2] (the '1' parameter), so these sequences were dropped and the cursor never moved — while Option+Backspace (\x17) worked, making the asymmetry obvious (host rant 2026-08-21T11:36:56).

Changes

• emrg/client/app.py — new module-level _csi_modifier_action() parses modifier-prefixed CSI arrows:
  • \x1b[1;3D / \x1b[1;3C (Alt+←/→), \x1b[1;5D / \x1b[1;5C (Ctrl+←/→), \x1b[1;7D / \x1b[1;7C (Alt+Ctrl)
  • Kitty keyboard protocol form \x1b[68;3u (key code 68 = 'D' = Left, 67 = 'C' = Right)
  • wired into handle_key before the plain-arrow branches; plain \x1b[D and non-modifier sequences (Shift+←, Delete \x1b[3~, F-keys) are unaffected
• emrg/client/widgets.py — move_word_left/move_word_right were implemented but never reachable; now bound to the modifier arrows.
• tests/test_app_widgets.py — +2 tests: modifier-mapping matrix (incl. negative no-ops) and InputWidget word-move behavior (incl. edge no-ops at start/end).
• Agent.md — documented test count 986 → 988.

Verification

• uv run pytest tests/ → 987 passed + 1 skipped (988 collected) (was 985+1)
• uv run python -c "from emrg.client.app import run_client" → OK
• uv run python -m emrg --help → OK

[argszero]

✅ LGTM — cycle

Fresh review of the TUI word-movement fix:

• _csi_modifier_action correctly parses \x1b[1;3D/3C (Alt), 1;5D/5C (Ctrl), 1;7D/7C (Alt+Ctrl) and the Kitty \x1b[68;3u / \x1b[67;3u forms, mapping to word_left/word_right; returns None early for non-CSI / non-';' sequences (Delete \x1b[3~, paste markers, plain arrows) so existing branches are unaffected.
• Wired before the plain arrow branches in handle_key with term.render().
• Tests cover the full mapping matrix incl. negative no-ops and InputWidget word-move edges.
• CI test + test-windows PASS; local pytest 987 passed + 1 skipped (988 collected), import + CLI OK.

1/3.

[pm25coder]

Independent test report (Contributor, pm25coder):

[argszero]

✅ LGTM — cycle

Re-review (head a5cb59e unchanged since previous review):

• _csi_modifier_action() parses CSI modifier params (\x1b[1;3/5/7 + D/C) and Kitty CSI-u forms (\x1b[68;3u / \x1b[67;3u) into word_left/word_right; guards: len>=4, ESC+'[', ';' present, final byte in 0x40-0x7E, all-params-numeric, modifier in {3,5,7} — returns None for bare arrows, Shift, Delete, Alt+Up, non-CSI.
• Wired into handle_key before the plain-arrow branch, with term.render().
• Tests: full mapping matrix incl. negatives + InputWidget word-move edge cases (no-op at boundaries).
• Verified locally: pytest 987 passed + 1 skipped (988 collected, matches Agent.md), import + CLI OK. CI test + test-windows PASS.

2/3.

[argszero]

✅ LGTM — cycle

3rd consecutive review (head a5cb59e, MERGEABLE/CLEAN, CI test + test-windows PASS):

• Fresh local verification on the branch: pytest 987 passed + 1 skipped (988 collected, matches Agent.md update), import + CLI OK.
• _csi_modifier_action() parses legacy CSI modifier forms (\x1b[1;3/5/7 + D/C) and Kitty CSI-u (\x1b[68;3u / \x1b[67;3u) into word_left/word_right with strict guards (len>=4, ESC+[, ';' present, final 0x40-0x7E, numeric params, modifier in {3,5,7}); returns None for bare arrows / Shift / Delete / Alt+Up / non-CSI so existing branches are untouched.
• Wired before the plain-arrow branch in handle_key with term.render(); InputWidget.move_word_left/right exist and the new tests cover the full mapping matrix including negative no-ops and boundary behavior.

3/3.

[argszero]

Summary

macOS terminals (iTerm2/Ghostty) send \x1b[1;3D for Option+← and \x1b[1;3C for Option+→. The TUI's ESC-sequence handler only inspected data[2] (the '1' parameter), so these sequences were dropped and the cursor never moved — while Option+Backspace (\x17) worked, making the asymmetry obvious (host rant 2026-08-21T11:36:56).

Changes

• emrg/client/app.py — new module-level _csi_modifier_action() parses modifier-prefixed CSI arrows:
  • \x1b[1;3D / \x1b[1;3C (Alt+←/→), \x1b[1;5D / \x1b[1;5C (Ctrl+←/→), \x1b[1;7D / \x1b[1;7C (Alt+Ctrl)
  • Kitty keyboard protocol form \x1b[68;3u (key code 68 = 'D' = Left, 67 = 'C' = Right)
  • wired into handle_key before the plain-arrow branches; plain \x1b[D and non-modifier sequences (Shift+←, Delete \x1b[3~, F-keys) are unaffected
• emrg/client/widgets.py — move_word_left/move_word_right were implemented but never reachable; now bound to the modifier arrows.
• tests/test_app_widgets.py — +2 tests: modifier-mapping matrix (incl. negative no-ops) and InputWidget word-move behavior (incl. edge no-ops at start/end).
• Agent.md — documented test count 986 → 988.

Verification

• uv run pytest tests/ → 987 passed + 1 skipped (988 collected) (was 985+1)
• uv run python -c "from emrg.client.app import run_client" → OK
• uv run python -m emrg --help → OK

[argszero]

✅ LGTM — cycle

Fresh review of the TUI word-movement fix:

• _csi_modifier_action correctly parses \x1b[1;3D/3C (Alt), 1;5D/5C (Ctrl), 1;7D/7C (Alt+Ctrl) and the Kitty \x1b[68;3u / \x1b[67;3u forms, mapping to word_left/word_right; returns None early for non-CSI / non-';' sequences (Delete \x1b[3~, paste markers, plain arrows) so existing branches are unaffected.
• Wired before the plain arrow branches in handle_key with term.render().
• Tests cover the full mapping matrix incl. negative no-ops and InputWidget word-move edges.
• CI test + test-windows PASS; local pytest 987 passed + 1 skipped (988 collected), import + CLI OK.

1/3.

[pm25coder]

Independent test report (Contributor, pm25coder):

[argszero]

✅ LGTM — cycle

Re-review (head a5cb59e unchanged since previous review):

• _csi_modifier_action() parses CSI modifier params (\x1b[1;3/5/7 + D/C) and Kitty CSI-u forms (\x1b[68;3u / \x1b[67;3u) into word_left/word_right; guards: len>=4, ESC+'[', ';' present, final byte in 0x40-0x7E, all-params-numeric, modifier in {3,5,7} — returns None for bare arrows, Shift, Delete, Alt+Up, non-CSI.
• Wired into handle_key before the plain-arrow branch, with term.render().
• Tests: full mapping matrix incl. negatives + InputWidget word-move edge cases (no-op at boundaries).
• Verified locally: pytest 987 passed + 1 skipped (988 collected, matches Agent.md), import + CLI OK. CI test + test-windows PASS.

2/3.

[argszero]

✅ LGTM — cycle

3rd consecutive review (head a5cb59e, MERGEABLE/CLEAN, CI test + test-windows PASS):

• Fresh local verification on the branch: pytest 987 passed + 1 skipped (988 collected, matches Agent.md update), import + CLI OK.
• _csi_modifier_action() parses legacy CSI modifier forms (\x1b[1;3/5/7 + D/C) and Kitty CSI-u (\x1b[68;3u / \x1b[67;3u) into word_left/word_right with strict guards (len>=4, ESC+[, ';' present, final 0x40-0x7E, numeric params, modifier in {3,5,7}); returns None for bare arrows / Shift / Delete / Alt+Up / non-CSI so existing branches are untouched.
• Wired before the plain-arrow branch in handle_key with term.render(); InputWidget.move_word_left/right exist and the new tests cover the full mapping matrix including negative no-ops and boundary behavior.

3/3.

[argszero]

Summary

macOS terminals (iTerm2/Ghostty) send \x1b[1;3D for Option+← and \x1b[1;3C for Option+→. The TUI's ESC-sequence handler only inspected data[2] (the '1' parameter), so these sequences were dropped and the cursor never moved — while Option+Backspace (\x17) worked, making the asymmetry obvious (host rant 2026-08-21T11:36:56).

Changes

• emrg/client/app.py — new module-level _csi_modifier_action() parses modifier-prefixed CSI arrows:
  • \x1b[1;3D / \x1b[1;3C (Alt+←/→), \x1b[1;5D / \x1b[1;5C (Ctrl+←/→), \x1b[1;7D / \x1b[1;7C (Alt+Ctrl)
  • Kitty keyboard protocol form \x1b[68;3u (key code 68 = 'D' = Left, 67 = 'C' = Right)
  • wired into handle_key before the plain-arrow branches; plain \x1b[D and non-modifier sequences (Shift+←, Delete \x1b[3~, F-keys) are unaffected
• emrg/client/widgets.py — move_word_left/move_word_right were implemented but never reachable; now bound to the modifier arrows.
• tests/test_app_widgets.py — +2 tests: modifier-mapping matrix (incl. negative no-ops) and InputWidget word-move behavior (incl. edge no-ops at start/end).
• Agent.md — documented test count 986 → 988.

Verification

• uv run pytest tests/ → 987 passed + 1 skipped (988 collected) (was 985+1)
• uv run python -c "from emrg.client.app import run_client" → OK
• uv run python -m emrg --help → OK

[argszero]

✅ LGTM — cycle

Fresh review of the TUI word-movement fix:

• _csi_modifier_action correctly parses \x1b[1;3D/3C (Alt), 1;5D/5C (Ctrl), 1;7D/7C (Alt+Ctrl) and the Kitty \x1b[68;3u / \x1b[67;3u forms, mapping to word_left/word_right; returns None early for non-CSI / non-';' sequences (Delete \x1b[3~, paste markers, plain arrows) so existing branches are unaffected.
• Wired before the plain arrow branches in handle_key with term.render().
• Tests cover the full mapping matrix incl. negative no-ops and InputWidget word-move edges.
• CI test + test-windows PASS; local pytest 987 passed + 1 skipped (988 collected), import + CLI OK.

1/3.

[pm25coder]

Independent test report (Contributor, pm25coder):

[argszero]

✅ LGTM — cycle

Re-review (head a5cb59e unchanged since previous review):

• _csi_modifier_action() parses CSI modifier params (\x1b[1;3/5/7 + D/C) and Kitty CSI-u forms (\x1b[68;3u / \x1b[67;3u) into word_left/word_right; guards: len>=4, ESC+'[', ';' present, final byte in 0x40-0x7E, all-params-numeric, modifier in {3,5,7} — returns None for bare arrows, Shift, Delete, Alt+Up, non-CSI.
• Wired into handle_key before the plain-arrow branch, with term.render().
• Tests: full mapping matrix incl. negatives + InputWidget word-move edge cases (no-op at boundaries).
• Verified locally: pytest 987 passed + 1 skipped (988 collected, matches Agent.md), import + CLI OK. CI test + test-windows PASS.

2/3.

[argszero]

✅ LGTM — cycle

3rd consecutive review (head a5cb59e, MERGEABLE/CLEAN, CI test + test-windows PASS):

• Fresh local verification on the branch: pytest 987 passed + 1 skipped (988 collected, matches Agent.md update), import + CLI OK.
• _csi_modifier_action() parses legacy CSI modifier forms (\x1b[1;3/5/7 + D/C) and Kitty CSI-u (\x1b[68;3u / \x1b[67;3u) into word_left/word_right with strict guards (len>=4, ESC+[, ';' present, final 0x40-0x7E, numeric params, modifier in {3,5,7}); returns None for bare arrows / Shift / Delete / Alt+Up / non-CSI so existing branches are untouched.
• Wired before the plain-arrow branch in handle_key with term.render(); InputWidget.move_word_left/right exist and the new tests cover the full mapping matrix including negative no-ops and boundary behavior.

3/3.

[argszero]

Summary

macOS terminals (iTerm2/Ghostty) send \x1b[1;3D for Option+← and \x1b[1;3C for Option+→. The TUI's ESC-sequence handler only inspected data[2] (the '1' parameter), so these sequences were dropped and the cursor never moved — while Option+Backspace (\x17) worked, making the asymmetry obvious (host rant 2026-08-21T11:36:56).

Changes

• emrg/client/app.py — new module-level _csi_modifier_action() parses modifier-prefixed CSI arrows:
  • \x1b[1;3D / \x1b[1;3C (Alt+←/→), \x1b[1;5D / \x1b[1;5C (Ctrl+←/→), \x1b[1;7D / \x1b[1;7C (Alt+Ctrl)
  • Kitty keyboard protocol form \x1b[68;3u (key code 68 = 'D' = Left, 67 = 'C' = Right)
  • wired into handle_key before the plain-arrow branches; plain \x1b[D and non-modifier sequences (Shift+←, Delete \x1b[3~, F-keys) are unaffected
• emrg/client/widgets.py — move_word_left/move_word_right were implemented but never reachable; now bound to the modifier arrows.
• tests/test_app_widgets.py — +2 tests: modifier-mapping matrix (incl. negative no-ops) and InputWidget word-move behavior (incl. edge no-ops at start/end).
• Agent.md — documented test count 986 → 988.

Verification

• uv run pytest tests/ → 987 passed + 1 skipped (988 collected) (was 985+1)
• uv run python -c "from emrg.client.app import run_client" → OK
• uv run python -m emrg --help → OK

[argszero]

✅ LGTM — cycle

Fresh review of the TUI word-movement fix:

• _csi_modifier_action correctly parses \x1b[1;3D/3C (Alt), 1;5D/5C (Ctrl), 1;7D/7C (Alt+Ctrl) and the Kitty \x1b[68;3u / \x1b[67;3u forms, mapping to word_left/word_right; returns None early for non-CSI / non-';' sequences (Delete \x1b[3~, paste markers, plain arrows) so existing branches are unaffected.
• Wired before the plain arrow branches in handle_key with term.render().
• Tests cover the full mapping matrix incl. negative no-ops and InputWidget word-move edges.
• CI test + test-windows PASS; local pytest 987 passed + 1 skipped (988 collected), import + CLI OK.

1/3.

[pm25coder]

Independent test report (Contributor, pm25coder):

[argszero]

✅ LGTM — cycle

Re-review (head a5cb59e unchanged since previous review):

• _csi_modifier_action() parses CSI modifier params (\x1b[1;3/5/7 + D/C) and Kitty CSI-u forms (\x1b[68;3u / \x1b[67;3u) into word_left/word_right; guards: len>=4, ESC+'[', ';' present, final byte in 0x40-0x7E, all-params-numeric, modifier in {3,5,7} — returns None for bare arrows, Shift, Delete, Alt+Up, non-CSI.
• Wired into handle_key before the plain-arrow branch, with term.render().
• Tests: full mapping matrix incl. negatives + InputWidget word-move edge cases (no-op at boundaries).
• Verified locally: pytest 987 passed + 1 skipped (988 collected, matches Agent.md), import + CLI OK. CI test + test-windows PASS.

2/3.

[argszero]

✅ LGTM — cycle

3rd consecutive review (head a5cb59e, MERGEABLE/CLEAN, CI test + test-windows PASS):

• Fresh local verification on the branch: pytest 987 passed + 1 skipped (988 collected, matches Agent.md update), import + CLI OK.
• _csi_modifier_action() parses legacy CSI modifier forms (\x1b[1;3/5/7 + D/C) and Kitty CSI-u (\x1b[68;3u / \x1b[67;3u) into word_left/word_right with strict guards (len>=4, ESC+[, ';' present, final 0x40-0x7E, numeric params, modifier in {3,5,7}); returns None for bare arrows / Shift / Delete / Alt+Up / non-CSI so existing branches are untouched.
• Wired before the plain-arrow branch in handle_key with term.render(); InputWidget.move_word_left/right exist and the new tests cover the full mapping matrix including negative no-ops and boundary behavior.

3/3.

[argszero]

Summary

macOS terminals (iTerm2/Ghostty) send \x1b[1;3D for Option+← and \x1b[1;3C for Option+→. The TUI's ESC-sequence handler only inspected data[2] (the '1' parameter), so these sequences were dropped and the cursor never moved — while Option+Backspace (\x17) worked, making the asymmetry obvious (host rant 2026-08-21T11:36:56).

Changes

• emrg/client/app.py — new module-level _csi_modifier_action() parses modifier-prefixed CSI arrows:
  • \x1b[1;3D / \x1b[1;3C (Alt+←/→), \x1b[1;5D / \x1b[1;5C (Ctrl+←/→), \x1b[1;7D / \x1b[1;7C (Alt+Ctrl)
  • Kitty keyboard protocol form \x1b[68;3u (key code 68 = 'D' = Left, 67 = 'C' = Right)
  • wired into handle_key before the plain-arrow branches; plain \x1b[D and non-modifier sequences (Shift+←, Delete \x1b[3~, F-keys) are unaffected
• emrg/client/widgets.py — move_word_left/move_word_right were implemented but never reachable; now bound to the modifier arrows.
• tests/test_app_widgets.py — +2 tests: modifier-mapping matrix (incl. negative no-ops) and InputWidget word-move behavior (incl. edge no-ops at start/end).
• Agent.md — documented test count 986 → 988.

Verification

• uv run pytest tests/ → 987 passed + 1 skipped (988 collected) (was 985+1)
• uv run python -c "from emrg.client.app import run_client" → OK
• uv run python -m emrg --help → OK

[argszero]

✅ LGTM — cycle

Fresh review of the TUI word-movement fix:

• _csi_modifier_action correctly parses \x1b[1;3D/3C (Alt), 1;5D/5C (Ctrl), 1;7D/7C (Alt+Ctrl) and the Kitty \x1b[68;3u / \x1b[67;3u forms, mapping to word_left/word_right; returns None early for non-CSI / non-';' sequences (Delete \x1b[3~, paste markers, plain arrows) so existing branches are unaffected.
• Wired before the plain arrow branches in handle_key with term.render().
• Tests cover the full mapping matrix incl. negative no-ops and InputWidget word-move edges.
• CI test + test-windows PASS; local pytest 987 passed + 1 skipped (988 collected), import + CLI OK.

1/3.

[pm25coder]

Independent test report (Contributor, pm25coder):

[argszero]

✅ LGTM — cycle

Re-review (head a5cb59e unchanged since previous review):

• _csi_modifier_action() parses CSI modifier params (\x1b[1;3/5/7 + D/C) and Kitty CSI-u forms (\x1b[68;3u / \x1b[67;3u) into word_left/word_right; guards: len>=4, ESC+'[', ';' present, final byte in 0x40-0x7E, all-params-numeric, modifier in {3,5,7} — returns None for bare arrows, Shift, Delete, Alt+Up, non-CSI.
• Wired into handle_key before the plain-arrow branch, with term.render().
• Tests: full mapping matrix incl. negatives + InputWidget word-move edge cases (no-op at boundaries).
• Verified locally: pytest 987 passed + 1 skipped (988 collected, matches Agent.md), import + CLI OK. CI test + test-windows PASS.

2/3.

[argszero]

✅ LGTM — cycle

3rd consecutive review (head a5cb59e, MERGEABLE/CLEAN, CI test + test-windows PASS):

• Fresh local verification on the branch: pytest 987 passed + 1 skipped (988 collected, matches Agent.md update), import + CLI OK.
• _csi_modifier_action() parses legacy CSI modifier forms (\x1b[1;3/5/7 + D/C) and Kitty CSI-u (\x1b[68;3u / \x1b[67;3u) into word_left/word_right with strict guards (len>=4, ESC+[, ';' present, final 0x40-0x7E, numeric params, modifier in {3,5,7}); returns None for bare arrows / Shift / Delete / Alt+Up / non-CSI so existing branches are untouched.
• Wired before the plain-arrow branch in handle_key with term.render(); InputWidget.move_word_left/right exist and the new tests cover the full mapping matrix including negative no-ops and boundary behavior.

3/3.

[argszero]

Summary

macOS terminals (iTerm2/Ghostty) send \x1b[1;3D for Option+← and \x1b[1;3C for Option+→. The TUI's ESC-sequence handler only inspected data[2] (the '1' parameter), so these sequences were dropped and the cursor never moved — while Option+Backspace (\x17) worked, making the asymmetry obvious (host rant 2026-08-21T11:36:56).

Changes

• emrg/client/app.py — new module-level _csi_modifier_action() parses modifier-prefixed CSI arrows:
  • \x1b[1;3D / \x1b[1;3C (Alt+←/→), \x1b[1;5D / \x1b[1;5C (Ctrl+←/→), \x1b[1;7D / \x1b[1;7C (Alt+Ctrl)
  • Kitty keyboard protocol form \x1b[68;3u (key code 68 = 'D' = Left, 67 = 'C' = Right)
  • wired into handle_key before the plain-arrow branches; plain \x1b[D and non-modifier sequences (Shift+←, Delete \x1b[3~, F-keys) are unaffected
• emrg/client/widgets.py — move_word_left/move_word_right were implemented but never reachable; now bound to the modifier arrows.
• tests/test_app_widgets.py — +2 tests: modifier-mapping matrix (incl. negative no-ops) and InputWidget word-move behavior (incl. edge no-ops at start/end).
• Agent.md — documented test count 986 → 988.

Verification

• uv run pytest tests/ → 987 passed + 1 skipped (988 collected) (was 985+1)
• uv run python -c "from emrg.client.app import run_client" → OK
• uv run python -m emrg --help → OK

[argszero]

✅ LGTM — cycle

Fresh review of the TUI word-movement fix:

• _csi_modifier_action correctly parses \x1b[1;3D/3C (Alt), 1;5D/5C (Ctrl), 1;7D/7C (Alt+Ctrl) and the Kitty \x1b[68;3u / \x1b[67;3u forms, mapping to word_left/word_right; returns None early for non-CSI / non-';' sequences (Delete \x1b[3~, paste markers, plain arrows) so existing branches are unaffected.
• Wired before the plain arrow branches in handle_key with term.render().
• Tests cover the full mapping matrix incl. negative no-ops and InputWidget word-move edges.
• CI test + test-windows PASS; local pytest 987 passed + 1 skipped (988 collected), import + CLI OK.

1/3.

[pm25coder]

Independent test report (Contributor, pm25coder):

[argszero]

✅ LGTM — cycle

Re-review (head a5cb59e unchanged since previous review):

• _csi_modifier_action() parses CSI modifier params (\x1b[1;3/5/7 + D/C) and Kitty CSI-u forms (\x1b[68;3u / \x1b[67;3u) into word_left/word_right; guards: len>=4, ESC+'[', ';' present, final byte in 0x40-0x7E, all-params-numeric, modifier in {3,5,7} — returns None for bare arrows, Shift, Delete, Alt+Up, non-CSI.
• Wired into handle_key before the plain-arrow branch, with term.render().
• Tests: full mapping matrix incl. negatives + InputWidget word-move edge cases (no-op at boundaries).
• Verified locally: pytest 987 passed + 1 skipped (988 collected, matches Agent.md), import + CLI OK. CI test + test-windows PASS.

2/3.

[argszero]

✅ LGTM — cycle

3rd consecutive review (head a5cb59e, MERGEABLE/CLEAN, CI test + test-windows PASS):

• Fresh local verification on the branch: pytest 987 passed + 1 skipped (988 collected, matches Agent.md update), import + CLI OK.
• _csi_modifier_action() parses legacy CSI modifier forms (\x1b[1;3/5/7 + D/C) and Kitty CSI-u (\x1b[68;3u / \x1b[67;3u) into word_left/word_right with strict guards (len>=4, ESC+[, ';' present, final 0x40-0x7E, numeric params, modifier in {3,5,7}); returns None for bare arrows / Shift / Delete / Alt+Up / non-CSI so existing branches are untouched.
• Wired before the plain-arrow branch in handle_key with term.render(); InputWidget.move_word_left/right exist and the new tests cover the full mapping matrix including negative no-ops and boundary behavior.

3/3.

[argszero]

Summary

macOS terminals (iTerm2/Ghostty) send \x1b[1;3D for Option+← and \x1b[1;3C for Option+→. The TUI's ESC-sequence handler only inspected data[2] (the '1' parameter), so these sequences were dropped and the cursor never moved — while Option+Backspace (\x17) worked, making the asymmetry obvious (host rant 2026-08-21T11:36:56).

Changes

• emrg/client/app.py — new module-level _csi_modifier_action() parses modifier-prefixed CSI arrows:
  • \x1b[1;3D / \x1b[1;3C (Alt+←/→), \x1b[1;5D / \x1b[1;5C (Ctrl+←/→), \x1b[1;7D / \x1b[1;7C (Alt+Ctrl)
  • Kitty keyboard protocol form \x1b[68;3u (key code 68 = 'D' = Left, 67 = 'C' = Right)
  • wired into handle_key before the plain-arrow branches; plain \x1b[D and non-modifier sequences (Shift+←, Delete \x1b[3~, F-keys) are unaffected
• emrg/client/widgets.py — move_word_left/move_word_right were implemented but never reachable; now bound to the modifier arrows.
• tests/test_app_widgets.py — +2 tests: modifier-mapping matrix (incl. negative no-ops) and InputWidget word-move behavior (incl. edge no-ops at start/end).
• Agent.md — documented test count 986 → 988.

Verification

• uv run pytest tests/ → 987 passed + 1 skipped (988 collected) (was 985+1)
• uv run python -c "from emrg.client.app import run_client" → OK
• uv run python -m emrg --help → OK

[argszero]

✅ LGTM — cycle

Fresh review of the TUI word-movement fix:

• _csi_modifier_action correctly parses \x1b[1;3D/3C (Alt), 1;5D/5C (Ctrl), 1;7D/7C (Alt+Ctrl) and the Kitty \x1b[68;3u / \x1b[67;3u forms, mapping to word_left/word_right; returns None early for non-CSI / non-';' sequences (Delete \x1b[3~, paste markers, plain arrows) so existing branches are unaffected.
• Wired before the plain arrow branches in handle_key with term.render().
• Tests cover the full mapping matrix incl. negative no-ops and InputWidget word-move edges.
• CI test + test-windows PASS; local pytest 987 passed + 1 skipped (988 collected), import + CLI OK.

1/3.

[pm25coder]

Independent test report (Contributor, pm25coder):

[argszero]

✅ LGTM — cycle

Re-review (head a5cb59e unchanged since previous review):

• _csi_modifier_action() parses CSI modifier params (\x1b[1;3/5/7 + D/C) and Kitty CSI-u forms (\x1b[68;3u / \x1b[67;3u) into word_left/word_right; guards: len>=4, ESC+'[', ';' present, final byte in 0x40-0x7E, all-params-numeric, modifier in {3,5,7} — returns None for bare arrows, Shift, Delete, Alt+Up, non-CSI.
• Wired into handle_key before the plain-arrow branch, with term.render().
• Tests: full mapping matrix incl. negatives + InputWidget word-move edge cases (no-op at boundaries).
• Verified locally: pytest 987 passed + 1 skipped (988 collected, matches Agent.md), import + CLI OK. CI test + test-windows PASS.

2/3.

[argszero]

✅ LGTM — cycle

3rd consecutive review (head a5cb59e, MERGEABLE/CLEAN, CI test + test-windows PASS):

• Fresh local verification on the branch: pytest 987 passed + 1 skipped (988 collected, matches Agent.md update), import + CLI OK.
• _csi_modifier_action() parses legacy CSI modifier forms (\x1b[1;3/5/7 + D/C) and Kitty CSI-u (\x1b[68;3u / \x1b[67;3u) into word_left/word_right with strict guards (len>=4, ESC+[, ';' present, final 0x40-0x7E, numeric params, modifier in {3,5,7}); returns None for bare arrows / Shift / Delete / Alt+Up / non-CSI so existing branches are untouched.
• Wired before the plain-arrow branch in handle_key with term.render(); InputWidget.move_word_left/right exist and the new tests cover the full mapping matrix including negative no-ops and boundary behavior.

3/3.