OpenCode Go: auto mode presents local estimates as authoritative when the cookie cache is empty

[Newarr]

Summary

In auto mode, OpenCode Go can show local estimates as if they were server values. The web overlay from #2430 only runs when a cookie is already cached. With an empty cache it is skipped without an error, the local strategy wins, and the menu bar shows quota percentages that are wrong by up to 55 points with no sign that they are local.

Not a duplicate of #2543 or #2551. Those covered the case where a cookie exists. This is the case where the overlay has no cookie to use.

Environment

CodexBar 0.49.6 (120), macOS 27.0, arm64. Provider opencodego, cookieSource auto, no workspaceID, local rows present in ~/.local/share/opencode/opencode.db.

Observed

Read 9 seconds apart, both normalized to percent used.

Window	Dashboard	CodexBar
5-hour	0%	0%
Weekly	100%	45%
Monthly	50%	0%

Reproduction

1. Use an account with local rows in opencode.db.
2. Start from an empty cookie cache, as on a fresh install.
3. Run auto mode. It succeeds with no warning:

$ codexbar usage --provider opencodego --source auto --format json
[{"provider":"opencodego","source":"local","usage":{
  "primary":  {"usedPercent":0,   "windowMinutes":300,  "resetsAt":"2026-08-16T18:17:41Z"},
  "secondary":{"usedPercent":45.4,"windowMinutes":10080,"resetsAt":"2026-08-16T23:59:59Z"},
  "tertiary": {"usedPercent":0,   "windowMinutes":43200,"resetsAt":"2026-09-15T20:55:55Z"}}}]

$ codexbar usage --provider opencodego --source web --format json
[{"source":"web","provider":"opencodego","error":{"kind":"provider",
  "message":"No OpenCode session cookies found in browsers.","code":1}}]

Cause

The state cannot heal itself. The overlay needs a cached cookie, and the only strategy that fills that cache never runs.

OpenCodeGoLocalUsageFetchStrategy.snapshot in OpenCodeGoProviderDescriptor.swift:

guard context.settings?.opencodego?.cookieSource != .off,
      let cookie = Self.cachedOrManualCookie(context: context)
else {
    return (snapshot, false)   // local only, no error
}
// ... (never a fresh browser import here).

resolveStrategies returns [local, web], and ProviderFetchPipeline moves to the next strategy only on a thrown error. A successful local read ends the pipeline, so the web strategy that imports the cookie is never reached.

Expected

Either let the overlay import a cookie when the cache is empty, or mark local-only results in the UI the way the CLI marks them with "source":"local". An estimate should not render as an authoritative quota.

Workaround

Set workspaceID in ~/.codexbar/config.json. Auto then puts the web strategy first and matches the dashboard. This appears only in a comment on closed issue #2543.

#2879 would remove the cookie dependency here entirely.

No cookie values, tokens, or account identifiers are included in this report.

[clawsweeper]

Codex review: keeping this open for maintainer follow-up; there is still a little grit to resolve. Reviewed August 16, 2026, 12:05 PM ET / 16:05 UTC.

Summary
This remains a distinct, source-proven OpenCode Go routing bug: an unscoped local success ends auto mode before it can obtain authoritative web windows. The linked API-key PR does not cover the no-cookie, no-key path, and the auth/privacy behavior needs maintainer direction.

Reproducibility: yes. at source level: unscoped auto mode returns after a successful local read, while the web strategy is the only path that can import a first cookie. Focused strategy tests establish the relevant ordering.

Root-cause cluster
Relationship: canonical
Canonical: #2982
Summary: This is the canonical cache-empty routing branch; related reports and PRs cover cached-cookie overlays, broader calculation symptoms, or API-key support.

Members:

• partial_overlap: fix: overlay authoritative OpenCode Go web windows onto local usage #2430 - It added an authoritative overlay only when a cached or manual cookie is already available.
• partial_overlap: OpenCode Go in auto mode only shows local data; real web usage never gets applied #2543 - It concerned the cached-cookie route, whereas this report identifies the cache-empty route.
• adjacent_distinct: OpenCode Go Usage calculation incorrect #2551 - It reports inaccurate usage calculations without establishing the cache-empty routing mechanism.
• adjacent_distinct: Add OpenCode Go usage API support #2879 - It adds an API-key path and does not demonstrate coverage for the reported no-cookie, no-key case.

Proposal only: this assessment does not dispatch repair, suppress jobs, mutate sibling items, close, or merge anything.

Maintainer decision needed

Question	Recommendation
Should unscoped OpenCode Go auto mode import a browser cookie after a successful local read, or retain local-first routing and visibly label local estimates?	Disclose local estimates: Keep local-first routing and add a visible estimate/source indicator wherever local quota windows are shown.

Why: Current docs intentionally avoid a fresh browser import on this path, while changing that behavior affects provider authentication and user privacy.

Next step
A focused repair is feasible, but maintainers must first choose between first-refresh browser import and clear local-estimate disclosure.

Review details

Best possible solution:

Adopt one explicit contract: preserve privacy-first local routing and visibly label estimates, or intentionally import a browser cookie in auto mode with focused privacy, fallback, and fresh-install coverage.

Do we have a high-confidence way to reproduce the issue?

Yes, at source level: unscoped auto mode returns after a successful local read, while the web strategy is the only path that can import a first cookie. Focused strategy tests establish the relevant ordering.

Is this the best way to solve the issue?

Unclear until maintainers choose the auth/privacy contract. Visible estimate disclosure is the narrowest repair because current documentation intentionally avoids first-refresh browser imports.

AGENTS.md: found and applied where relevant.

Remaining risk / open question:

• Allowing a first-refresh browser import changes the documented provider-auth and privacy behavior; retaining local-only output without a visible distinction continues to make estimates easy to mistake for authoritative quotas.

Codex review notes: model internal, reasoning high; reviewed against 355f8443a55c.

Label changes

Label changes:

• add P2: Incorrect quota windows are a clear but limited-impact problem for OpenCode Go users.
• add impact:auth-provider: The behavior depends on cookie availability and automatic provider-source routing.
• add impact:ux-friction: Users can mistake estimates for authoritative quotas, and the available workaround requires manual configuration.
• add issue-rating: 🦞 diamond lobster: Current issue advisory state selects this label.
• add clawsweeper:source-repro: Current issue advisory state selects this label.
• add clawsweeper:no-new-fix-pr: Current issue advisory state selects this label.
• add clawsweeper:needs-maintainer-review: Current issue advisory state selects this label.
• add clawsweeper:needs-product-decision: Current issue advisory state selects this label.

Label justifications:

• P2: Incorrect quota windows are a clear but limited-impact problem for OpenCode Go users.
• impact:auth-provider: The behavior depends on cookie availability and automatic provider-source routing.
• impact:ux-friction: Users can mistake estimates for authoritative quotas, and the available workaround requires manual configuration.

Evidence reviewed

What I checked:

• Local-first route: Unscoped auto mode explicitly orders the local strategy before the web strategy. (Sources/CodexBarCore/Providers/OpenCodeGo/OpenCodeGoProviderDescriptor.swift:108, 355f8443a55c)
• Successful strategy is terminal: The shared fetch pipeline returns immediately after a strategy succeeds; it reaches another strategy only after a fallback-eligible error. (Sources/CodexBarCore/Providers/ProviderFetchPlan.swift:321, 355f8443a55c)
• Empty-cache behavior: The local strategy returns its snapshot when no cached or manual cookie exists, and the code documents that this overlay path never performs a fresh browser import. (Sources/CodexBarCore/Providers/OpenCodeGo/OpenCodeGoProviderDescriptor.swift:189, 355f8443a55c)
• Documented and tested boundary: Current docs and focused tests describe local-first unscoped auto mode, web-first scoped exceptions, and authoritative overlays only with cached or manual cookies. (docs/opencode.md:32, 355f8443a55c)
• Feature history: Available history identifies a recent OpenCode Go descriptor contributor; the cached-cookie overlay itself was merged as the related PR below. (Sources/CodexBarCore/Providers/OpenCodeGo/OpenCodeGoProviderDescriptor.swift:108, a15887deddd9)
• Related open PR: The supplied PR context says the API-key path retains local-first automatic discovery, so it cannot demonstrably solve the reported cache-empty path when no API key is configured. (2f42b83b74a7)

Likely related people:

• aseelye: Authored the merged overlay work that defines the current cached/manual-cookie boundary. (role: introduced cached-cookie overlay behavior; confidence: high; commits: b036579b4b05; files: Sources/CodexBarCore/Providers/OpenCodeGo/OpenCodeGoProviderDescriptor.swift, Tests/CodexBarTests/OpenCodeGoWebOverlayTests.swift)
• kentoku.matsunami: Most recent available history for the OpenCode Go descriptor identifies this contributor. (role: recent area contributor; confidence: medium; commits: a15887deddd9; files: Sources/CodexBarCore/Providers/OpenCodeGo/OpenCodeGoProviderDescriptor.swift)
• Peter Steinberger: Merged the related cached-cookie overlay PR into current feature history. (role: merger of related feature work; confidence: medium; commits: b036579b4b05; files: Sources/CodexBarCore/Providers/OpenCodeGo/OpenCodeGoProviderDescriptor.swift)

How this review workflow works

• ClawSweeper keeps one durable marker-backed review comment per issue or PR.
• Re-runs edit this comment so the latest verdict, findings, and automation markers stay together instead of adding duplicate bot comments.
• A fresh review can be triggered by eligible @clawsweeper re-review comments, exact-item GitHub events, scheduled/background review runs, or manual workflow dispatch.
• PR/issue authors and users with repository write access can comment @clawsweeper re-review or @clawsweeper re-run on an open PR or issue to request a fresh review only.
• Maintainers can also comment @clawsweeper review to request a fresh review only.
• Fresh-review commands do not start repair, autofix, rebase, CI repair, or automerge.
• Maintainer-only repair and merge flows require explicit commands such as @clawsweeper autofix, @clawsweeper automerge, @clawsweeper fix ci, or @clawsweeper address review.
• Maintainers can comment @clawsweeper explain to ask for more context, or @clawsweeper stop to stop active automation.