Skip to content

Releases: omerakben/aha

AHA v1.0.0 — agent-native single-door kit

02 Jun 16:10
ec6493e

Choose a tag to compare

AHA v1.0.0 — agent-native, single-door alignment kit.

Live: https://omerakben.github.io/aha/

  • /aha is the single door: it runs the full cycle (ask → align → critique → optimize → handoff) and detects resume / debrief / rule-promotion from .aha/state.yaml. No new command names.
  • Private session memory in .aha/state.yaml (gitignored by default); promoted rules land in AGENTS.md / CLAUDE.md only after an explicit patch-preview + approval.
  • Typed leverage questions, an alignment ledger (confidence + basis + would-change-if), premortem critique, and an eval rubric.
  • Agent-to-agent handoff_contract plus a mandatory return_packet; the debrief flags self-contradicting reports as conflicts instead of trusting them.
  • Safe self-evolution: conflict scan, supersedes, review dates, one rule per debrief.
  • The five standalone skills (/ask-me, /align-me, /critique-this, /optimize-prompt, /debrief) stay first-class for quick single-step checks. Zero runtime dependencies.

Install (Claude Code): /plugin marketplace add omerakben/aha then /plugin install aha@aha. Cursor/manual: clone and run ./install.sh.