Greptile Summary

This PR introduces light-terminal support to the TUI theme layer by adding palette auto-detection (OPENCLAW_THEME env var + COLORFGBG heuristic), a WCAG AA-targeted lightPalette, and a matching createSyntaxTheme light branch. The dark theme is completely unchanged. The implementation is clean and follows the module-level evaluation pattern already used elsewhere in the codebase.

Key observations:

• Detection logic (isLightBackground) is correct — explicit OPENCLAW_THEME takes precedence over COLORFGBG, and the default preserves dark behavior.
• The 7 new tests are well-structured, using vi.resetModules() + dynamic import to correctly exercise module-init-time evaluation.
• One color in the light syntax theme — #FF0000 used for attr and attribute tokens in syntax-theme.ts (lines 35-36) — has a contrast ratio of ~4.0:1 against white, which falls below the WCAG AA threshold of 4.5:1 claimed by the PR. Using #C50000 would satisfy the requirement at ~4.9:1 contrast.

Confidence Score: 4/5

• Safe to merge with a minor color fix. The dark theme is unchanged and the light theme is opt-in only.
• The PR is well-structured with sound logic and thorough tests. The dark theme is entirely unchanged and light theme is opt-in, so there is no behavioral risk. One light-mode syntax color (#FF0000 for attr/attribute) does not meet the WCAG AA 4.5:1 contrast target the PR explicitly guarantees. This is a straightforward fix (change to #C50000), but should be corrected before merge to fulfill the PR's stated accessibility claim.
• src/tui/theme/syntax-theme.ts — Fix #FF0000 to #C50000 for lines 35-36 to meet WCAG AA 4.5:1 contrast threshold.

_{Last reviewed commit: 87b5921}

@ademczuk screenshot before or after or im closing this PR as i have no way of verifying without a build

CI Status

The 3 failing checks are pre-existing type errors on main, unrelated to this PR:

• check — TS errors in src/discord/monitor/message-handler.preflight.test.ts, src/discord/monitor/message-handler.queue.test.ts, src/line/bot-handlers.test.ts, src/telegram/bot-native-commands.session-meta.test.ts
• checks (extensions) — extensions/diffs/src/http.ts:180 TypeError in hasProxyForwardingHints
• secrets — fork PRs can't access repo secrets (expected)

This PR only touches src/tui/theme/ (3 files). All relevant checks pass:

• checks (node, test) ✅
• checks (bun, test) ✅
• checks-windows (all 6 shards) ✅
• build-artifacts ✅
• install-smoke ✅
• dead-code report ✅

Aisle Security Finding — Addressed

Fixed in a35aa35. The COLORFGBG parsing now:

1. Length cap — rejects values over 64 bytes (real COLORFGBG is 3-7 chars like "0;15")
2. lastIndexOf instead of split — extracts only the last segment without allocating an array
3. Range clamp 0..255 — rejects out-of-range values and implicitly rejects NaN

This addresses CWE-400 (uncontrolled resource consumption) by bounding both input size and output range before any computation.

need verification of changes

Rendered using the exact hex values from src/tui/theme/theme.ts darkPalette and lightPalette on Windows 11. Three panels:

Every color shown corresponds 1:1 to the palette objects in the diff. The isLightBackground() function selects lightPalette automatically when OPENCLAW_THEME=light or COLORFGBG indicates a light background; dark users see zero change.

Pushed 50e71a2 with one more fix - using the actual xterm RGB lookup table for 256-colour cube luminance instead of raw 0-5 indices. The previous approximation misclassified some bright backgrounds (e.g. bright cyan at index 39) as dark.

Summary of changes across all commits:

1. isLightBackground() detection - checks OPENCLAW_THEME env var first (explicit override), then parses COLORFGBG terminal variable for 16-colour palette, 256-colour greyscale ramp, and 6x6x6 colour cube
2. Dual darkPalette/lightPalette - dark palette is unchanged from the original, light palette uses high-contrast colours for white/light backgrounds
3. Light syntax highlighting theme - VS Code light-inspired, WCAG AA compliant (#C50000 for attrs after greptile caught the original #FF0000)
4. 19 tests covering all detection branches

Verification steps:

# Run tests
pnpm vitest run src/tui/theme/theme.test.ts

# Manual - force light mode
OPENCLAW_THEME=light openclaw gateway status

# Manual - auto-detect from terminal
COLORFGBG=0;15 openclaw gateway status

All 10 bot comments from chatgpt-codex-connector and greptile addressed across 5 rounds. CI green (secrets failure is expected for fork PRs).

@vincentkoc these CI checks could do with some intent engineering, as this should be a slamdunk of a PR.

@vincentkoc - verification of the changes:

What changed (commit 50e71a2):
The 256-colour cube branch in isLightBackground() was using raw indices (0-5) for BT.601 luminance instead of actual xterm RGB values. Fixed by adding the xterm lookup table [0, 95, 135, 175, 215, 255] and computing luminance on real RGB values (0-255 scale) with threshold 128.

Before: lum = 0.299 * r + 0.587 * g + 0.114 * b where r/g/b were 0-5 indices. Threshold 2.5.
After: lum = 0.299 * rVal + 0.587 * gVal + 0.114 * bVal where rVal/gVal/bVal are actual RGB channel values. Threshold 128.

Verification:

• All 19 existing tests pass without modification (the threshold change was calibrated correctly)
• bg=231 (white, RGB 255,255,255, lum=255) → light ✅
• bg=16 (black, RGB 0,0,0, lum=0) → dark ✅
• bg=244 (greyscale ramp, ≥50% grey) → light ✅
• bg=232 (near-black greyscale) → dark ✅
• Full cube sweep: 141 light / 75 dark classifications match expected luminance distribution

No other files changed. Only 8 lines modified in theme.ts (the cube luminance block).

pnpm vitest run src/tui/theme/theme.test.ts  # 19 tests pass
pnpm tsgo                                     # zero errors

CI: all test jobs green. check failure is an upstream regression affecting all open PRs (confirmed on #39238, #39245).

Pushed a rebased update to fix/tui-color-contrast-light-themes.

What changed:

• switched 256-color cube auto-detection to pick the palette by actual text contrast instead of a lum >= 128 midpoint, which addresses the last live Codex thread for bright 256-color backgrounds like COLORFGBG=15;34 and 15;39
• darkened the light-theme structural colors so borders / blockquote rails / code fences are readable on light terminals
• kept the light syntax theme and WCAG-safe red attr colors
• documented OPENCLAW_THEME=light|dark in the TUI docs and env-var reference
• added focused tests for the new detection cases plus explicit AA contrast assertions for the light palette

Verification:

• pnpm vitest run src/tui/theme/theme.test.ts

The branch is rebased onto current origin/main, the review threads on the TUI theme files are resolved, and the PR is assigned to vincentkoc.

@vincentkoc - thanks for the rewrite, the contrast-ratio detection is a much better approach than the fixed threshold.

I've added the missing changelog entry (a5f3808). All CI was green on 5e1b6dc and the branch is rebased on current main.

Since you authored the core rewrite commit (5e1b6dc TUI: harden light theme contrast), cleanest landing path is probably cherry-pick to main and close - happy to have it landed however works best. If you'd rather squash-merge the PR instead, the CHANGES_REQUESTED review would need dismissing or a fresh approval.

Closing this PR because it looks dirty (too many unrelated or unexpected changes). This usually happens when a branch picks up unrelated commits or a merge went sideways. Please recreate the PR from a clean branch.

Closing this PR because it looks dirty (too many unrelated or unexpected changes). This usually happens when a branch picks up unrelated commits or a merge went sideways. Please recreate the PR from a clean branch.

Superseded by #40339 - same branch, CHANGELOG conflict resolved, clean diff against main (6 files).

This PR was auto-closed after the branch was force-pushed and couldn't be reopened.

Too many files changed for review. (3000 files found, 500 file limit)

Branch got contaminated with upstream commits during rebase - ended up with a 3k file diff that's obviously unreviewable. Closing to cut down on review queue noise. The original colour contrast fix was a 2-file change, but it's not worth salvaging at this point given upstream terminal rendering has changed since.