Greptile Summary

This PR adds workspace-based agent inference to the TUI startup flow. When openclaw tui is launched from inside a configured agent workspace directory, it automatically selects that agent as the active context — without requiring any explicit --session agent:<id>:... flag.

Key implementation details:

• resolveAgentIdsByWorkspacePath / resolveAgentIdByWorkspacePath in agent-scope.ts normalize both the current working directory and configured workspace paths consistently via fs.realpathSync.native() (with lexical fallback for non-existent paths), then compare specificity using path length.
• resolveInitialTuiAgentId in tui.ts enforces a clear priority: explicit agent-prefixed session keys win first, then falls through to workspace inference, then to the default fallback agent.
• All returned agent IDs are normalized (via listAgentIds → normalizeAgentId), ensuring consistency across all three code paths.
• Tests comprehensively cover specificity ordering, no-match fallback, symlink alias resolution via realpath, explicit session override behavior, and TUI startup inference logic.

No issues found. Path normalization is applied consistently to both sides of every comparison. The explicit-session-wins contract is properly enforced before workspace inference is attempted. The workspace-inferred agent ID is already normalized and gracefully handled by the existing TUI state reconciliation in case the inferred agent is not present in the live server agent list.

Confidence Score: 5/5

• This PR is safe to merge — the workspace inference logic is well-isolated, properly tested, and does not introduce any functional bugs or regressions.
• The workspace-based agent inference is a low-risk, well-scoped feature. Path normalization is consistent and delegates properly to OS-level calls for cross-platform robustness. The priority order in resolveInitialTuiAgentId is enforced correctly (explicit > inferred > default). All code paths produce normalized agent IDs, and the TUI state reconciliation gracefully handles agents that may not be present on the live server. Tests comprehensively verify the critical invariants including specificity ordering, no-match fallback, symlink resolution, and the explicit-override path.
• No files require special attention.

_{Last reviewed commit: 2968436}

@codex review
@greptileai review

Codex Review: Something went wrong. Try again later by commenting “@codex review”.

We were unable to download your code in a timely manner.

Merged via squash.

• Prepared head SHA: 23533e24c4cbb91fd5c3006485c9a5d69bd5c2fb
• Merge commit: 492fe679a703a0809be1575c65183af02a0a4182

Thanks @arceus77-7!