Repository navigation
Description
Summary
I ran an audit over the open bug backlog to find reports that later commits appear to have quietly fixed without the issue being closed. This lists 15 candidates for closing, each with the commit that appears to fix it and one line of current-code evidence so you can verify a claim in a few seconds rather than taking my word for it.
Nothing here is a new bug report. If this kind of triage isn't useful, say so and I won't file more.
Audit pinned to a0a654c3a9 (v1.4.178-rc.2). Line numbers are as of that commit.
Method and limits — please read before acting
- Static analysis only. Nothing here was reproduced at runtime, on any platform.
- Candidates were found by matching each open bug against commit subjects landed after it was filed, then investigated by reading the actual diffs and the code at
HEAD. - Every claim below then went through an adversarial pass that specifically tried to refute it — checking that the cited code says what was claimed, that the commit diff (not its subject) addresses the defect, and that the failure mode doesn't survive in a parallel path (macOS/Windows/Linux branch, SSH/remote twin, folder-workspace path).
- That refutation pass killed 18 of ~50 initial claims — roughly a third. In one case the supposed fix landed a day before the report was filed. The 15 below are the survivors, but treat that error rate as the prior: these are "appears fixed, worth a look", not "definitely fixed, close it."
- This is AI-assisted analysis, disclosed per the spirit of the AI-review requirement in
CONTRIBUTING.md.
Coverage: 129 of the ~541 open bug issues — the ones with the strongest textual match to a later commit. The remainder weren't investigated, so absence from this list means nothing.
Higher confidence (11)
| Issue | Reported symptom | Appears fixed by | Evidence at a0a654c3a9 |
|---|---|---|---|
| #1715 | GitHub Projects uses the wrong gh host in multi-host setups; false "missing project scope" |
971b167548 (#8932) |
project-view/internals.ts:33-40 — projectGhExecOptions(host) pins host per call; git/runner.ts:1512 injects --hostname at spawn |
| #9193 | terminal close --tab reports not-found for live tabless PTYs that pane-close can kill |
817197fc31 (#9288) |
orca-runtime.ts:27157 — if (!tabId) { return this.closeTerminal(handle) }. The terminal_tab_not_found throw site is gone from the codebase |
| #9553 | Issues panel runs an unscoped global search for self-hosted git: projects |
63da958813 (#9668) |
github/client.ts:1108-1110 — repo:owner/repo now pushed unconditionally; ownerRepo typed non-null at :1100 |
| #9873 | Crash reports lost: bundle 502, then JSON fallback 404s from api.onorca.dev |
746a8f3e48 (#10023) |
ipc/feedback.ts:17 — fallback constant deleted; retries hit primary. No api.onorca.dev anywhere in src/ |
| #8629 | Windows hook lifecycle test can overwrite the caller's live Codex hooks | e58de71f5e (#9501) |
managed-hook-stdin-lifecycle.test.ts:16-22 — beforeEach scopes ORCA_USER_DATA_PATH to an mkdtempSync fixture, restored in afterEach |
| #8950 | Onboarding raises computerUsePermissions:openSetup IPC error when the macOS helper app is missing |
9ced27eca8 (#8951) |
onboarding-feature-setup.ts:283 — helperUnavailableReason short-circuits before openComputerUsePermissionSetup() |
| #8258 | Cursor Agent CLI not detected on Windows (manual PowerShell launch) | 4c7bbed2fb (#8266) |
agent-node-entrypoint-identities.ts:10-16 — exact-path match for cursor-agent/versions/<v>/index.js |
| #10367 | Cmd+J palette: ArrowDown can't cross from Recent Worktrees into Open Tabs | e6e197feed (#13076) |
worktree-palette-create-action.ts:38 — SELECTABLE_ENTRY_TYPES now includes workspace-tab, simulator-tab, project-target |
| #8075 | Markdown editor lacks numeric superscript footnotes | ac8879c789 (#8307) |
rich-markdown-html-superscript-link.ts:20 — dedicated <sup><a href> node; merged one day after the report |
| #9652 | "Link Routing" description stays English under Korean UI | 6bf34421d2 (#9444) |
browser-link-routing-copy.ts:29 — routed through translate() with {{shortcut}}; ko/es/ja/zh entries added |
| #9442 | Link Routing description not localized | 6bf34421d2 (#9444) |
Same fix — #9442 and #9652 are duplicates of each other |
Lower confidence (4) — needs someone who knows the subsystem
| Issue | Reported symptom | Appears fixed by | Note |
|---|---|---|---|
| #8038 | Terminal drops last Hangul syllable onto a new line | fe6f929c6e (#11293) |
The confirmed repro in-thread is Shift+Enter, which never reaches xterm's CompositionHelper — it's intercepted at keyboard-handlers.ts:608-628. The fix is the deferred-newline path, covered by korean-ime-terminal-shift-enter-commit.spec.ts. Reporter's build predates it. Wants a native-IME confirmation |
| #8272 | Remote Windows host disappears while smart sort refreshes it | 84bd306949 (#13662) |
worktree-sort-order-update.ts:20-26 short-circuits no-op sort writes. Mechanically verified; the user-visible symptom is inferred, not reproduced |
| #9331 | Resuming an OMP session from history opens with Pi identity | e3adb20917 (#8991) + f655ae3929 (#9600) |
agent-session-resume.ts:16 — 'omp' added to RESUMABLE_TUI_AGENTS (absent at 057a81493a, the commit before the report) |
| #10273 | Mobile Relay stays unavailable after a transient desktop broker failure | 0da5380a49 (#10147) + fd72a8926 (#10260) |
relay-auth-coordinator.ts:210-217 — retryable errors now scheduleRetry() with backoff instead of parking at offline |
One correction to flag
#11598 should stay open, and it's worth a label. b86880a9c1 (#12214) says Closes #11598 but only fixed half of it: the manual-sleep discard is genuinely gone, while auto-hibernation still refuses to sleep any pane whose last run was interrupted — agent-hibernation-planner.ts:147-148 bails on entry.interrupted === true with no expiry, so interrupt-heavy workflows still accumulate resident agent processes. A maintainer already noted this in-thread.
Also found
16 further issues are partially resolved — headline symptom fixed, a named sub-defect still live. The recurring pattern is a fix applied to the local path but not its SSH/WSL/remote twin. I've left those off since they're still legitimately open, but I'm happy to post the list if it's useful for scoping.
I can also extend the audit to the ~300 open bugs that weren't covered.
Activity
AmethystLiang commentedon Aug 12, 2026
thanks for reporting. looking into it
nwparker commentedon Aug 31, 2026
Audit follow-up (2026-08-30): the ten higher-confidence open candidates were verified against merged fixes and closed with solution links: #1715, #9193, #9873, #8629, #8950, #8258, #10367, #8272, #9331, plus duplicate #9442. Lower-confidence candidates #8038, #11598, and other partial cases were left untouched where the original report still has a live or unverified facet.
Open PRs whose linked issues were already fixed or explicitly superseded were also closed with replacement/duplicate links. The audit remains open for the broader backlog rather than claiming unverified fixes.
nwparker commentedon Aug 31, 2026
btw thx for the report!
I'll close for now since we'll track down the others outside since they're lower confidence 🫡