Skip to content

v9.0.2

Choose a tag to compare

@github-actions github-actions released this 19 Sep 22:13

[9.0.2] - 2026-09-19

Same-goal follow-ups continue more often; inspect-then-implement still stops.

  • Manager skills now treat method or emphasis narrowing, extra evidence, and
    "do the research and save the plan" as continuation of the active goal.
    Inspect-only followed by implementation, mixed extra scope, and a replacement
    goal still require an explicit continue, defer, or abandon choice.
  • The Jev alignment corpus adds those continue cases and retunes the same-goal
    score so live measurement is 8/8. Jev remains eval-only and is not in the
    plugin.
  • Offline feature qualification may pin a newer Bun than 8.3.0. Dependencies
    and engines still must match the live baseline.
  • Session v5 schema, validation gates, reviewer authority, and continuation
    routing are unchanged. No live model eval or paid canary was run for this
    candidate.

Upgrade with opencode plugin opencode-plugin-flow@9.0.2 --global --force.

Reviewed offline feature qualification

Policy: alignment-examples-v1, for 9.0.2 only. Reviewed code: 206e380.
Candidate artifact: sha256:9a94e39a9331b064283a7174a5f042d10b4f2d65236c25f0bc3d88eb4e960133. Qualification record: sha256:6f5689fc8c781e081d55ea4fce0462c18f99b57989433a568ad8f7054db4d2e8.
No live model evals or canary were run for 9.0.2. Reviewed code, deterministic checks, replay and provider-free OpenCode UI/configuration checks support this feature release; they do not establish its live model behavior.
Prior measured artifact (8.3.0): sha256:f2cad40877d616a25df9f675ddbd4aeb0e0a6a342ef246e5d109a3978c891563.
Prior bundle: sha256:61eac17e755f95216eed58a1361eceb72a6dcd59c94c8f325cd0bc8d1958075f. Prior canary: sha256:8de004b8718714f73b361695d0cc54c423c711064ee79decf5d4d233e5950bd6.
Prior results only: 76/76. These are not measurements of 9.0.2, and no release after 8.3.0 has been measured live.
Scope approval: Operator approval in this OpenCode session, 2026-09-19. Explicitly authorized offline qualification for 9.0.2 alignment-examples-v1, after being told a funded live campaign was the alternative and that this path does not measure live model behavior. The candidate ships manager same-goal continue examples and an eval-only Jev corpus retune. Offline feature qualification may pin bun 1.4.0 while 8.3.0 remains the live baseline. No new tool, command, setting, or Session v5 field. The last live-measured release remains 8.3.0. No new live model eval or canary was authorized or run.