github Gentleman-Programming/gentle-ai v2.1.8

latest release: v2.1.9
23 hours ago

What's New

Gentle AI v2.1.8 unblocks every reported review-lifecycle dead-end: stale SDD verify routing, corrupted or legacy authority inventories, next-slice and recovery-chain gate denials, first publications to empty remotes, and silently lost reviewer results. Native review now recovers explicitly from every one of these states instead of failing closed forever.

Changes

Authority Recovery and Repair

  • Audited reclaim for incomplete stores: gentle-ai review reclaim quarantines incomplete compact store entries behind two-phase audit records instead of leaving the inventory undiscoverable. (#1400)
  • Invalid recovery-edge reconciliation: gentle-ai review reconcile-authority repairs invalid unchanged-target recovery edges through a sole-anomaly, evidence-bound ceremony that preserves history byte-for-byte. (#1429)
  • Pre-contract authorization repair: the same ceremony now repairs historical escalated-recovery edges whose free-form maintainer authorization predates the exact v1 binding, unbricking 2.1.6-era inventories without weakening validation. (#1436)
  • Scoped corruption discovery: authority corruption is confined to candidate authorities instead of poisoning unrelated lineages during discovery. (#1391)
  • Legacy inventories stay readable: historical zero_edit_escalation records and receiptless terminal legacy chains load read-only as historical instead of failing the whole store. (#1397, #1403)

Gate Routing and Delivery

  • Next-slice workspaces route to fresh review: new work on top of an approved committed receipt classifies as unrelated or scope-changed instead of poisoning gate evaluation, with git infrastructure failures surfaced typed instead of silently reclassified. (#1431)
  • Chained recovery deliveries bind at publication: scope-changed recovery successors created after a committed delivery now compose their approved predecessor chain and rebind at pre-push/pre-pr, with every failure path keeping the original denial. (#1432)
  • Compatible base advance at pre-pr: current-changes receipts reconcile against an advanced base instead of denying scope-changed. (#1413)
  • Pure scope contraction accepted: correction recovery accepts deliveries that only shrink the reviewed scope. (#1428)
  • Zero-edit corrections escalate terminally: an unchanged tree with a failing original and passing regression transitions to terminal escalation instead of looping. (#1423)
  • First publication to empty remotes: pre-push bootstrap validates the delivered range against immutable genesis scope and requires pre-base ancestry to be disclosed by the reviewed base tree, uniformly across target kinds and both native and compact gates. (#1437)
  • Publication range deduplication: repeated approved paths across a reviewed range no longer false-invalidate delivery. (#1425)

Reviewer Result Integrity

  • START emits binding inputs: orchestrators build capture bindings exclusively from START's target_identity and lens_bindings; the bundled plugin forwards full lens names instead of rewriting them. (#1395, #1412)
  • Capture preflight and incident preservation: the plugin preflights each capture binding before launching a reviewer, honors GENTLE_AI_REVIEW_CWD for nested repositories, and preserves failed results as replayable incident artifacts via gentle-ai review preserve-result. (#1434)

SDD and Agent Integration

  • Stale verify routes to verify: compact authority discovery binds stale verify evidence to the selected change and preserves remediation routing. (#1386)
  • Claude orchestrator parity: the Claude Code SDD workflow carries the same session-preflight and entry-routing hard gates as OpenCode. (#1427)
  • Codex default permissions: gentle-dev injection now cleans up legacy configuration safely instead of stacking custom permission blocks. (#1416)
  • Malformed status artifacts rejected: SDD status rejects non-object and malformed review transaction JSON instead of misreporting. (#1390)
  • Typed git process-control failures: process-start failures carry typed causes across platforms, including the Windows exec-lookup case. (#1406, #1426)

Upgrade

brew upgrade gentle-ai && gentle-ai sync

Links

Don't miss a new gentle-ai release

NewReleases is sending notifications on new releases.