fix: stop branch-reconcile probing origin on a repo that has none

[atomantic]

Summary

The server log was emitting an unactionable error on a loop:

❌ branch-reconcile: git ls-remote origin failed — remote branch state unknown this cycle

Root cause. gatherBranchState resolves origin and computes hasOrigin, then fired git ls-remote --heads origin in its read fan-out regardless of that verdict. reconcile() gates both of its own remote reads on hasOrigin precisely because "probing one every cycle would log a failure forever" — the gather never got that gate.

The caller that made it visible is the read-only leftover-branch detector (userActionDetectors.js:75), which calls gatherBranchState once for every managed app. An app's repoPath can be a directory that was never a clone (a static-site folder, a checkout whose remote was never added), so ls-remote exited 128 there on every detector cycle — for a repo with no remote to read and zero candidate branches to judge. gatherBranchState doesn't throw on a non-repo (it returns 0 inputs), so the detector's own try/catch never fired and nothing named the offending app.

What changed

• gatherBranchState gates the remote read on the hasOrigin it already has — the same gate reconcile() applies at both of its remote-reading sites. null there is the established "could not ask" value that reconcile() already passes explicitly for an origin-less repo, so the sole consumer (upstreamGone) is unaffected: it requires false for both senses of null, and an origin-less repo has no origin/* upstream to begin with. Fail-safe direction — nothing reads as "shipped", nothing gets reaped.
• The failure line now names the repo and git's own reason. One log line serves three call sites across every managed app; a bare "ls-remote failed" told you neither which repo went unread nor whether it was a network blip, an unauthenticated remote, or no remote at all:

  ❌ branch-reconcile: git ls-remote origin failed in <repo> (fatal: 'origin' does not appear to be a git repository) — remote branch state unknown this cycle
  

  A rejected execGit (30s timeout, spawn failure) and a non-zero exit previously collapsed into the same message; they are now distinguishable.

Test plan

• Two new tests, each verified failing before the fix:
  • gatherBranchState "skips the remote probe on a repo with no origin" — asserts no ls-remote spawn and no error log (failed pre-fix: ls-remote was called).
  • reapOrphanedRemotes "names the repo and git's own reason when the remote cannot be read" (failed pre-fix: the line contained neither).
• server/services/branchReconcile.test.js + userActionDetectors.test.js: 156 passed.
• Full npm test (server + client): 1908 files / 38506 tests passed, 0 failures.
• Verified against the live instance: calling gatherBranchState on the real non-git app path no longer logs, PortOS's own gather still resolves its branches, and a direct listRemoteHeads call still reports — now naming the repo and the cause.

https://claude.ai/code/session_01FBgckWpLV6qzytK9vKnVvU