Skip to content

Gitmoot v0.7.0

Latest
Latest

Choose a tag to compare

View all tags
@jerryfane jerryfane released this 27 Jun 19:50
· 1 commit to main since this release
v0.7.0
06b9cdd
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

Headline: the complete automatic template-optimization loop for SkillOpt, plus runtime-reliability and orchestration improvements. Every new SkillOpt capability is off by default, additive, and keeps promotion manual.

SkillOpt — automatic template optimization (new, off by default)

Agent templates can now self-improve from real usage, feeding the existing gitmoot-skillopt optimizer — no weight training, manual promotion preserved.

  • Mode A — outcome harvest (#465/#468): verifiable job outcomes (merge/CI/review/revert) become auto-trace feedback; automatic revert detection flips earlier positives (#467); deterministic tool checkers (dupl/jscpd/golangci-lint/gocyclo/diff-size) add objective dimensions (#485).
  • Cross-family review signal (#469/#470): a different-model-family reviewer scores quality + scope-fidelity as a soft, secondary signal.
  • Mode B — champion/challenger (#473): Thompson-sampling bandit + gitmoot skillopt ab, live-traffic A/B interception for ask agents (#482), and a cross-family LLM-judge auto-pairwise (#483).
  • Promotion (#471/#472): configurable auto-promote with a confidence guardrail + candidate.* notifications; canary + auto-rollback (#484) routes a sampled fraction of traffic to a canary and rolls back on regression.
  • Judge hardening (epic #344): cross-family judge jury with median/majority/minority-veto + disagreement flag (#349); live pairwise evaluation — a blinded paired review packet (gitmoot-skillopt v0.4.0) ingested into canonical feedback (#508). (Trajectory digest for the judge ships in gitmoot-skillopt v0.4.0.)

Runtime reliability

  • Claude transient-401 ("socket connection closed") retry with exponential backoff (#487/#509).
  • Bounded re-ask loop on malformed agent output instead of a hard failure (#495).
  • More reliable read-only agent ask (no implement-guard block, clear timeout) (#496).

Orchestration & dashboard

  • --recipe flag for deterministic fan-out (#477).
  • Delegation worktree + branch cleanup (#492); job-kill releases stranded locks and terminalizes queued children (#491).
  • Dashboard surfaces in-flight jobs as an active_jobs view (#505).

Tests

  • Full-chain E2Es: canary lifecycle (#504), Mode A harvest (#465), and a cross-repo pairwise round-trip (#514).

Pairs with gitmoot-skillopt v0.4.0. Upgrade: gitmoot update.

Assets 7
Loading