Fix Claude probe session pollution

[steipete]

Summary

• persist one installation-local Claude probe UUID in the existing ClaudeProbe working directory
• pass that UUID through --session-id for both direct and watchdog-backed Claude CLI launches
• remove stale probe-owned transcripts before launch and keep the identifier file permission-restricted

Root cause

The CLI usage source starts an interactive Claude session so it can invoke /usage and /status. It previously launched without a session identity, causing every cold probe launch to register another empty account-side session.

Claude CLI 2.1.212 still has no non-interactive quota command. Its --no-session-persistence flag is print-mode-only, so it cannot drive the interactive /usage panel. Reusing an explicit session ID preserves every field the existing probe collects without creating a new identity per refresh.

Background delegated refresh is already policy-gated after #2191, so this currently affects user-initiated CLI refreshes rather than scheduled background probes. Repeated refreshes and cold launches still polluted the account session list.

Verification

• swift test --filter ClaudeCLISessionTests — 3 passed
• swift test --filter ClaudeDirectUsageFallbackTests — 3 passed
• swift test --filter ClaudeProbeWorkingDirectoryTests — 6 passed
• make check — SwiftFormat clean; SwiftLint 0 violations; repository checks passed
• autoreview — clean, no accepted/actionable findings

Live account-side counting was unavailable: the working local Claude CLI 2.1.206 is logged out, while the global 2.1.212 wrapper is missing its native optional dependency. No credentialed probe was attempted.

Closes #2251

[clawsweeper]

Codex review: needs maintainer review before merge. Reviewed July 17, 2026, 2:20 PM ET / 18:20 UTC.

Summary
The PR persists an installation-local Claude probe UUID, passes it as --session-id through direct and watchdog-backed Claude CLI launches, cleans stale probe transcripts, and adds session-ID persistence tests.

Reproducibility: yes. in source terms: the report identifies the Claude CLI usage probe launch path that previously lacked a stable --session-id, and the branch adds tests around the replacement behavior. A high-confidence live authenticated reproduction was not available because no credentialed Claude CLI probe was run.

Review metrics: 2 noteworthy metrics.

• Patch scope: 2 files affected; 116 added, 3 removed. The implementation is tightly limited to the Claude launch path and its focused regression suite.
• Focused coverage: 3 named test filters reported; 12 focused tests. The PR reports coverage for session reuse, direct fallback behavior, and probe working-directory handling.

Merge readiness
Overall: 🐚 platinum hermit
Proof: 🌊 off-meta tidepool
Patch quality: 🐚 platinum hermit
Result: ready for maintainer review.

Overall follows the weaker of proof and patch quality, so missing proof can cap an otherwise strong patch.

Rank-up moves:

• Let the pending build and macOS test checks complete.
• [P2] When an authenticated Claude CLI environment is available, add a redacted before/after verification that repeated probes reuse one account-side session.

Risk before merge

• [P1] Focused tests validate local UUID persistence and argument construction, but the PR body does not include a redacted authenticated Claude account-side observation confirming that repeated real probes create no additional session entries.

Maintainer options:

1. Decide the mitigation before merge
   Land the stable probe-session identity after normal maintainer review and completed checks, then obtain a redacted live verification when a working authenticated Claude CLI environment is available.
2. Pause or close
   Do not merge this PR until maintainers decide whether the risk is worth taking.

Next step before merge

• [P2] Repository policy requires explicit human handling for an owner-authored PR; no narrow automated repair is indicated.

Maintainer decision needed

• Question: Should this owner-authored Claude CLI behavior change be approved after the pending repository checks complete, with live account-side confirmation treated as follow-up evidence rather than a merge gate?
• Rationale: Repository policy keeps owner-authored pull requests open for explicit maintainer judgment; the remaining uncertainty is production-account behavior that unit tests cannot observe.
• Likely owner: steipete — Repository-owner authorship and the directly affected Claude probe implementation make this the clearest available decision route.
• Options:
  • Approve after checks (recommended): Merge the narrow tested fix once required checks pass, and collect redacted authenticated probe evidence when an environment is available.
  • Require live confirmation first: Wait for a redacted account-side before/after count showing repeated probes reuse one Claude session before approval.

Security
Cleared: The patch only adds local UUID persistence, restricted file permissions, cleanup of probe-owned artifacts, and CLI argument forwarding; no dependency, workflow, permission, or secret-handling expansion is indicated.

Review details

Best possible solution:

Land the stable probe-session identity after normal maintainer review and completed checks, then obtain a redacted live verification when a working authenticated Claude CLI environment is available.

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

Yes in source terms: the report identifies the Claude CLI usage probe launch path that previously lacked a stable --session-id, and the branch adds tests around the replacement behavior. A high-confidence live authenticated reproduction was not available because no credentialed Claude CLI probe was run.

Is this the best way to solve the issue?

Yes, provisionally: reusing one private installation-local session identifier preserves the existing interactive usage probe while avoiding a fresh account-side session per cold launch. A future non-interactive Claude quota interface could be preferable if the CLI provides one, but the PR body says none is currently available for this workflow.

AGENTS.md: found and applied where relevant.

Codex review notes: model internal, reasoning high; reviewed against 7ea74d4ccf12.

Label changes

Label changes:

• add P2: This fixes recurring provider-specific account session pollution with a bounded user-impact scope.
• add rating: 🐚 platinum hermit: Overall readiness is 🐚 platinum hermit; proof is 🌊 off-meta tidepool and patch quality is 🐚 platinum hermit.
• add status: 👀 ready for maintainer look: ClawSweeper has no concrete contributor-facing blocker left for this PR. Not applicable: This is a maintainer-authored PR, so external-contributor real-behavior proof is not a merge gate; its reported evidence is focused tests and checks, while live authenticated account-side confirmation remains unavailable.

Label justifications:

• P2: This fixes recurring provider-specific account session pollution with a bounded user-impact scope.
• rating: 🐚 platinum hermit: Overall readiness is 🐚 platinum hermit; proof is 🌊 off-meta tidepool and patch quality is 🐚 platinum hermit.
• status: 👀 ready for maintainer look: ClawSweeper has no concrete contributor-facing blocker left for this PR. Not applicable: This is a maintainer-authored PR, so external-contributor real-behavior proof is not a merge gate; its reported evidence is focused tests and checks, while live authenticated account-side confirmation remains unavailable.

Evidence reviewed

What I checked:

• Probe launch implementation: The branch prepares the probe working directory, removes stale probe-owned session artifacts, loads or creates a stable UUID, and supplies that UUID in the Claude launch arguments for both direct and watchdog-backed launches. (Sources/CodexBarCore/Providers/Claude/ClaudeCLISession.swift:299, 2d60bbb82665)
• Focused regression coverage: The added test suite verifies persisted UUID reuse, exact --session-id launch arguments, restricted identifier-file permissions, invalid-ID replacement, and fallback behavior. (Tests/CodexBarTests/ClaudeCLISessionTests.swift:1, 2d60bbb82665)
• Reported root cause and candidate fix: The linked report documents one account-side empty session per Claude usage probe; this PR explicitly closes that report with stable session reuse rather than changing the interactive /usage and /status probe workflow. (2d60bbb82665)
• Current review baseline: The PR is a cleanly mergeable two-file change against current main, with 116 additions and 3 deletions; remaining CI jobs were still in progress or queued at review time. (7ea74d4ccf12)

Likely related people:

• steipete: The PR is authored by the repository owner and changes the Claude CLI probe implementation under active maintenance. (role: repository owner and current area decision owner; confidence: high; commits: 2d60bbb82665; files: Sources/CodexBarCore/Providers/Claude/ClaudeCLISession.swift, Tests/CodexBarTests/ClaudeCLISessionTests.swift)
• Yuxin-Qiao: Merged work on the immediately adjacent Claude background-probe policy established the surrounding launch and refresh boundary relevant to this change. (role: adjacent Claude probe contributor; confidence: medium; commits: 5f136af0899f; files: Sources/CodexBarCore/Providers/Claude/ClaudeCLISession.swift)

What the crustacean ranks mean

• 🦀 challenger crab: rare, exceptional readiness with strong proof, clean implementation, and convincing validation.
• 🦞 diamond lobster: very strong readiness with only minor maintainer review expected.
• 🐚 platinum hermit: good normal PR, likely mergeable with ordinary maintainer review.
• 🦐 gold shrimp: useful signal, but proof or patch confidence is still limited.
• 🦪 silver shellfish: thin signal; proof, validation, or implementation needs work.
• 🧂 unranked krab: not merge-ready because proof is missing/unusable or there are serious correctness or safety concerns.
• 🌊 off-meta tidepool: rating does not apply to this item.

Shiny media proof means a screenshot, video, or linked artifact directly shows the changed behavior. Runtime, network, CSP, and security claims still need visible diagnostics.

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.

[steipete]

Merged as cfb7b8036087cc1186b963ada7bec2ce9efa71e2 from exact head 2d60bbb8266505652f5d9bb2542e8f910070140d.

Verification:

• swift test --filter ClaudeCLISessionTests: 3 tests passed.
• swift test --filter ClaudeDirectUsageFallbackTests: 3 tests passed.
• swift test --filter ClaudeProbeWorkingDirectoryTests: 6 tests passed.
• make check: passed; SwiftFormat 0/1535 files need formatting, SwiftLint 0 violations across 1534 files, and all repository checks passed.
• git diff --check: clean.
• Autoreview: clean; no accepted/actionable findings.
• CI on exact head: changes, lint, Linux x64, Linux arm64, macOS shards 0/2 and 1/2, aggregate lint-build-test, and GitGuardian all passed. macOS shards completed in 36m42s and 30m06s.
• Merge state before merge: CLEAN and MERGEABLE. No approval was required by repository policy.

Live claude.ai session-list counting could not be run from this worktree: the usable installed Claude CLI reports loggedIn: false, while the newer global wrapper is missing its native optional package. This limitation was documented before merge. Source inspection and tests verify every direct/watchdog probe launch now receives the same persisted UUID through --session-id, with a stable process-local fallback when persistence is unavailable.