Distillation

Date: Sep 15, 2026

• 🟡 [requested-review] (01:34) User requested a frozen, read-only adversarial correctness review of the complete V2 interrupted-session recovery candidate in /home/byk/Code/opencode.
• 🔴 [enforced-read-only-review] (01:34) User prohibited editing files, running formatters, mutating Git, touching production data, or starting services during the review.
• 🔴 (01:34) User specified exact review base commit ab9408c81cc9916177641ae8e9f08624fc27b5b9.
• 🔴 (01:34) User stated the committed head is dc661f30cb2055ad6248fdc59b14998d834497be plus an uncommitted validation delta.
• 🔴 (01:34) User identified /tmp/opencode/session-recovery-full.patch as the canonical complete candidate patch, with SHA-256 f051a53df476dd924bda0a14c4cd91a63c9e7eafff53bbe71f0460571335fcbd.
• 🔴 (01:34) User required reviewing every file in /tmp/opencode/session-recovery-full.patch, treating that patch as immutable scope and using current files only to obtain line numbers.
• 🔴 (01:34) User identified the built SEA as packages/opencode/dist/server/opencode-server, with SHA-256 1ecc9afe36c12b31a8892cdbfb8a1560ceec97255c3537c073b795351233be40.
• 🟡 (01:34) User asked to verify the canonical patch hash before reviewing.
• 🟡 (01:34) User requested review focus on: durable lease acquisition, takeover, and fencing; heartbeat and expiry races; explicit interruption and scope shutdown; recovery transaction atomicity, idempotence, and replay; newest-assistant supersession; ready/safe/unknown semantics; provider and local/hosted tool ambiguity; overflow compaction; pending inbox behavior; startup composition; migrations for fresh and upgraded databases; and regression quality.
• 🔴 (01:34) User established invariant: never call legacy SessionPrompt.
• 🔴 (01:34) User established invariant: never silently re-execute old calls.
• 🔴 (01:34) User established invariant: uncertain provider/tool outcomes become model-visible unknown errors.
• 🔴 (01:34) User established invariant: never recover an older assistant projection.
• 🔴 (01:34) User established invariant: EventV2 replay ownership stays separate.
• 🔴 (01:34) User established invariant: Core never imports Server.
• 🔴 (01:34) User established invariant: replace local ownership with durable multi-node ownership when clustered.
• 🔴 (01:34) User stated validation already observed: Core typecheck; schema, server, and opencode typechecks; 112 focused tests; full Core suite with 1108 tests; migration check; recorded cassette replay; Node and SEA builds; and isolated authenticated health, migration, and SIGTERM smoke testing.
• 🔴 (01:34) User stated the new regressions were applied to base dc661f30cb2055ad6248fdc59b14998d834497be and failed there with exactly 6 failures across the focused ownership/recovery files, then passed on the candidate.
• 🟡 (01:34) User requested a substantive report containing PASS, CONCERN, and MUST-FIX findings with exact current file:line evidence.
• 🔴 (01:34) User required the review output to end exactly MERGE or DO-NOT-MERGE; empty or progress-only output is invalid.
• 🔴 (01:36) User supplied project Effect guidance requiring current Effect v4 / effect-smol source and repository-local patterns rather than memory or older Effect v2/v3 examples.
• 🔴 (01:36) User specified that if .opencode/references/effect-smol is missing, https://github.com/Effect-TS/effect-smol should be cloned there in the project, not in the skill folder; Effect-specific work should search that source and nearby repository code first.
• 🔴 (01:36) User specified Effect implementation conventions: use Effect.gen(function* () { ... }) for multi-step workflows; Effect.fn("Name") or Effect.fnUntraced(...) for reusable service methods or important workflows; Effect Schema, branded schemas, and Schema.TaggedErrorClass for typed domain errors; thin HTTP handlers; explicit layer composition; and Effect-aware platform abstractions where consistent with surrounding code.
• 🔴 (01:36) User specified not to introduce any, non-null assertions, unchecked casts, or older Effect APIs merely to satisfy types.
• 🔴 [requested-tests] (01:36) User specified Effect testing patterns: use testEffect(...) from packages/opencode/test/lib/effect.ts for Effect services, layers, runtime context, scoped resources, or platform integrations; use it.live(...) for filesystem, Git repositories, HTTP servers, sockets, child processes, locks, real time, and other live platform behavior; prefer explicit test layers; and use scoped fixtures/finalizers for temporary directories, flags, databases, fibers, servers, and global state.
• 🔴 [enforced-workflow] (01:36) User directed: never run package tests from the repo root; run them from package directories such as packages/opencode.
• 🟡 (01:36) Tool output confirmed /tmp/opencode/session-recovery-full.patch has SHA-256 f051a53df476dd924bda0a14c4cd91a63c9e7eafff53bbe71f0460571335fcbd, exactly matching the user-specified canonical patch hash.
• 🟡 (01:36) Tool output confirmed packages/opencode/dist/server/opencode-server has SHA-256 1ecc9afe36c12b31a8892cdbfb8a1560ceec97255c3537c073b795351233be40, exactly matching the user-specified SEA hash.