Greptile Summary

This PR makes Control UI asset path resolution symlink-aware, fixing a regression where bun-installed global wrappers (which are symlinks like ~/.bun/bin/openclaw → .../dist/index.js) caused the dashboard to return 404 because directory-based heuristics evaluated the wrapper path instead of the real entrypoint.

Key changes:

• resolveControlUiDistIndexPath: builds an entrypointCandidates array by also resolving argv1 via fs.realpathSync, then runs Case 1 and the fallback package-boundary traversal against all candidates
• resolveControlUiRootSync: computes argv1RealpathDir from fs.realpathSync(argv1) and injects two extra candidate paths (<realpathDir>/dist/control-ui and <realpathDir>/control-ui) when the realpath differs from the nominal path
• Two regression tests cover both code paths for the symlinked Bun-style wrapper case

The implementation correctly handles symlink resolution with try/catch guards and proper multi-candidate fallback logic.

Confidence Score: 4/5

• Safe to merge; logic is sound, error cases are properly caught, and regression tests pass.
• The implementation correctly handles symlink resolution with appropriate error boundaries. The fallback logic properly supports multiple candidate paths, and the change from hard returns to breaks ensures all candidates are tried. Both affected functions have comprehensive test coverage for the symlinked wrapper scenario. The design is intentional and well-tested.
• No files require special attention; both changed files are straightforward and well-tested.

_{Last reviewed commit: ef6ab03}

AI-generated review note.

This looks partial, not a full fix for the reported 404 regressions.

The PR improves Control UI asset discovery for symlinked Bun/global wrappers in src/infra/control-ui-assets.ts, but the explicit-gateway.controlUi.root failure mode reported in #39621 still bypasses this path entirely.

The remaining likely failure path is deeper in file-open validation / boundary checks when serving assets from package-store layouts (pnpm/global-store hardlink-heavy installs), not just argv1 resolution.

So this may be OK as a narrow wrapper-discovery improvement, but it should not be treated as the fix for #39621 / #39693 without an HTTP-level regression test that proves GET / works when gateway.controlUi.root points at a global install path.

Thanks for calling this out — agreed the first revision was too narrow for the full #39621/#39693 surface.

I pushed a follow-up in commit 62424b6 that targets the explicit gateway.controlUi.root regression path as well.

What changed:

• handleControlUiHttpRequest now allows hardlinked files only when the resolved UI root is an OpenClaw package root (.../dist/control-ui with package name openclaw).
• Non-package roots keep the current hardlink rejection behavior.

Added regression coverage:

• src/gateway/control-ui.http.test.ts
  • rejects hardlinked index.html for non-package roots
  • allows hardlinked index.html for OpenClaw package roots
• src/gateway/server.control-ui-root.test.ts
  • configures gateway.controlUi.root to a pnpm/global-style path with hardlinked index.html
  • asserts GET / returns 200 and serves the dashboard HTML

This now exercises the explicit-root HTTP path directly, not only argv1 discovery.

Follow-up to unblock CI:

• rebased this branch onto latest openclaw/main
• refreshed generated apps/macos/Sources/OpenClaw/HostEnvSecurityPolicy.generated.swift in 66429ef

This addresses the failing check job (check:host-env-policy:swift reported the generated file as out-of-date).

Addressed the failing secrets check by refreshing .secrets.baseline in follow-up commit 98c57cbcc.

This is only a baseline line-number sync from detect-secrets (no functional code changes).

Rebased this branch onto latest main and force-pushed as b20f9e6b4 to resolve merge conflicts (DIRTY state).

The only conflict was .secrets.baseline; I regenerated/staged it with detect-secrets during rebase. No functional code changes in this rebase update.

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 #40385.

Barnacle closed this branch because the prep push picked up unrelated repo-wide changes. I recreated the Control UI fix on a clean branch from current main and kept only the intended 8-file payload:

• symlink-aware bundled asset discovery for global wrappers
• bundled-vs-resolved root classification for hardlink handling
• regression coverage for bundled auto-roots vs explicit configured roots

The replacement PR is here: #40385