v0.3.3

0.3.3 - 2026-10-29

• Stop placeholder-recovery polls from retaining unrelated comment bodies while preserving event ordering, pending placeholders, and the full comment API.
• Tell authors of stale PR reviews to comment @clawsweeper re-review, since later pushes are not reviewed automatically.
• Allow credential-free Team work-session provenance links on RoboClaw and GitHub-verified OpenClaw organization PRs without weakening privacy or proof requirements. Thanks @brokemac79.
• Restore scheduled audits, which failed on every run since 2026-10-19 because the workflow token was rejected as an invalid audit read token.
• Require one fresh final whole-branch review before publication or handoff, with optional early feedback and fresh review after relevant changes; retain separate proof, ClawSweeper, CI, and landing gates.
• Keep issue close proposals open when a human locks the issue after review. Thanks @brokemac79.
• Keep local committed-range contributor proof requirements independent of Git author display names, without treating local names as verified bot or Team identities.
• Skip unusable polling cache repairs without webhook coverage, and honor equal-timestamp comment deletions so stale bodies do not remain protected.
• Defer exact reviews the model provider throttles instead of retrying them as failures every few minutes.
• Prepare target-bound permission acquisition for review authority without changing active admission, transport retries, or caches.
• Confirm the human-review hold before removing existing stop-command protection, and fail safely instead of acknowledging failed or unverified label changes.
• Measure webhook payload bytes eligible for retirement in the bounded, read-only storage census.
• Count current exact and scheduled review runs in the dashboard's worker error rate, which was reporting 0% because it sampled none of them.