Skip to content

v1.31.0 — Loop engineering (LE-1..11)

Choose a tag to compare

View all tags
@feci feci released this 24 Jun 14:34
v1.31.0
1cd91e7
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

Four-tier loop-engineering program — designed by a real multi-agent Parley deliberation and implemented tier-by-tier with full multi-agent refutation review.

  • Tier 1 — verification honesty (LE-1/2/3/4): refutation-default Phase 6 review; strict_gate close veto; reviewer model-diversity signal; checks: pre-close command.
  • Tier 2 — loop budgets (LE-5/6/10): per-run driver budget (--max-driver-steps/--max-wall-clock, central [defaults.loop]) + escalate-on-breach; cost telemetry; status: candidate remediation ideas.
  • Tier 3 — close-decision integrity (LE-7/11): reservations / <2 reviewers escalate under auto_implement; fail-open 2-min goal-done gate before close.
  • Tier 4 — the outer loop (LE-8/9): COOPERATION.md §14 human brake + parley loop tick (one-shot, disabled-by-default, candidate-only). Hardened over 4 fix-up cycles vs frontmatter injection, digest collision, poisoned-dir liveness, symlink escape at any depth, and line-break separators.

Full changelog: see CHANGELOG.md. go test -count=1 ./... (incl. drift guard) green.

🤖 Generated with Claude Code

Assets 8
Loading