Skip to content

v2.2.3-rc.1 — Compact SDD attempts (community testing)

Pre-release
Pre-release

Choose a tag to compare

View all tags
@Alan-TheGentleman Alan-TheGentleman released this 30 Jul 00:10
v2.2.3-rc.1
cb8ffd1

Compact SDD runtime attempts — community prerelease

This prerelease reduces the context consumed by native SDD runtime-attempt control without removing its safety guarantees.

What changed

  • Added compact sdd-attempt acquire and sdd-attempt settle operations for normal sdd-apply, sdd-verify, and remediation orchestration.
  • Replaced growing RuntimeStatus payloads in generated agent guidance with bounded results: proceed, blocked, or complete, plus only the opaque continuation data required by the next operation.
  • Preserved the legacy status, begin, finish, and reset operations for diagnostics, scripts, and recovery.
  • Kept request idempotency, CAS protection, attempt/line budgets, immutable history, review bindings, and explicit-only reset behavior.
  • Added recovery for mutations that committed HEAD before reporting a publication failure.
  • Added exact replay/conflict handling for caller-supplied remediation evidence.

What we need tested

  1. Run a real SDD change through apply and verify. Confirm the agent no longer dumps the full attempt history into context.
  2. Interrupt an attempt after acquire, then resume. Confirm the active attempt is reported without launching duplicate work.
  3. Replay acquire or settle with the same operation-specific request ID. Confirm it converges idempotently.
  4. Confirm acquire and settle use different request IDs.
  5. Confirm reset is never executed automatically and still requires an explicit maintainer decision.
  6. If you use review remediation, exercise both an existing failed-evidence revision and the initial empty-evidence case.

Please include your OS, agent/client, gentle-ai --version, and the compact result returned by acquire/settle in reports.

Install

Download the binary for your platform from the assets below, verify it against SHA256SUMS.txt, and run it directly.

Example for Linux amd64:

sha256sum --check SHA256SUMS.txt
chmod +x gentle-ai_2.2.3-rc.1_linux_amd64
./gentle-ai_2.2.3-rc.1_linux_amd64 --version

These community-testing assets are unsigned. The checksum file provides integrity but not publisher authentication. This prerelease does not update Homebrew.

Verification status

  • Exact commit: cb8ffd186e83f780c390324da48ef615a7941efe
  • Required CI is green, including Windows Runtime, Darwin Runtime, Windows organic E2E, Linux organic E2E, unit tests, formatting, and Linux E2E lanes.
  • Local full tests, vet, format checks, focused compact-attempt tests, and golden tests passed.
  • The optional Windows Full Suite remains red on pre-existing Windows path/timeout cases that reproduce on the parent commit; the new compact-attempt tests pass in its Windows shards.

Known lifecycle issue

During validation we found a separate review-authority dead end after failed correction verification. It is tracked and ready for contribution in #1996. It is not hidden or treated as approved by this prerelease.

Assets 8
Loading