Distillation

Date: Sep 16, 2026

• 🔴 (13:31) User-provided WorkerInitData.forceWasm contract in packages/core/src/embedding-worker-types.ts: it is set only when respawning after a native worker posts init-needs-wasm; default/undefined prefers native. It is a no-op in vendored/SEA mode because that path uses __LORE_ORT_BINDING_PATH__ and never sets forceWasm, and it is also a no-op in dev/test because raw .ts has no sibling WASM.
• 🟡 (13:31) Assistant rejected the first independent audit as approval evidence because Lore context retrieval failed and it returned no findings; planned to finish admission-invariant checks locally and then run a fresh immutable review.
• 🔴 (13:32) The working-copy candidate tmzmtlss / Git commit e6d0f1edc850, description fix(core): correct embedding runtime selection and admission, is based on parent pxrvwsvv / 2cbf40060c67 on main, description feat(cch): add seeds for Claude Code up to 2.1.272 (#1780).
• 🔴 (13:32) Candidate scope is exactly 18 files with 1,818 insertions and 90 deletions: added packages/core/src/cgroup-memory.ts and packages/core/test/cgroup-memory.test.ts; modified packages/core/src/embedding/local.ts, packages/core/src/embedding/pool.ts, packages/core/src/embedding-cap.ts, packages/core/src/embedding-worker-types.ts, packages/core/src/embedding-worker.ts, packages/core/src/embedding.ts, packages/core/src/ort-native.ts, packages/core/test/embedding-cap.test.ts, packages/core/test/embedding-pool-memory.test.ts, packages/core/test/embedding-pool.test.ts, packages/core/test/embedding-worker-types.test.ts, packages/core/test/ort-native.test.ts, packages/gateway/script/bundle.ts, packages/gateway/script/ort-platform-package.ts, packages/gateway/test/bundle-exports.test.ts, and packages/gateway/test/ort-platform-package.test.ts.
• 🔴 (13:32) Working copy had no source conflicts; unrelated bookmark fix/responses-private-references had a conflict and remained untouched.
• 🟡 (13:32) Assistant planned to audit memory-admission math and the cgroup parser against stated invariants, prove each guard, and then run the full validation matrix.
• 🔴 (13:32) Project AGENTS.md states tests use a temporary SQLite database through packages/core/test/setup.ts, never the production database.
• 🔴 (13:32) Project release policy: never manually bump versions or edit CHANGELOG.md; Sentry Craft handles both automatically. Release workflow can be triggered with gh workflow run release.yml -f version=auto or an explicit version such as gh workflow run release.yml -f version=0.23.0.
• 🔴 (13:32) Release workflow ordering: 1. scripts/bump-version.sh updates all package.json versions; 2. changelog is generated from conventional commits; 3. a release/X.Y.Z branch is created and pushed; 4. CI tests and builds npm tarballs, standalone binaries, and delta patches; 5. Craft opens a publish issue; 6. labeling it accepted runs publishing through npm OIDC trusted publishing and creates the GitHub Release with binaries and patches.
• 🔴 (13:33) Root package.json pins packageManager to pnpm@10.28.0, Volta Node to 24.16.0, Volta pnpm to 10.28.0, and overrides onnxruntime-node to 1.27.0.
• 🔴 (13:33) Root validation scripts are pnpm test (vitest run, preceded by gateway bundling), pnpm run typecheck, pnpm run lint (oxlint --type-aware), pnpm run format:check (oxfmt --check), and pnpm run build; mutation testing uses stryker run.
• 🔴 (13:33) CONTRIBUTING.md was absent at /home/byk/Code/opencode-lore-embedding-runtime/CONTRIBUTING.md.
• 🔴 [enforced-workflow] (13:33) Project regression discipline requires every defect found by adversarial review to receive a deterministic same-PR regression that fails on the base branch, passes on the fix, and drives the real precondition; guard tests must fail when the guard is removed. The stated principle is: “Never trust the fix without seeing it fail.”
• 🔴 [requested-review] (13:33) quality/REVIEW.md requires two distinct review passes for non-trivial changes: 1. adversarial/correctness review covering break-inducing states, enumerated cases, apply-then-revert mutants, property-test stability ×10, hash-equivalence proofs, frozen-array mutation safety, and stale-cache analysis; 2. security/pentest review covering injection, credential leaks, protocol-level attacks, and privilege escalation. Adversarial review is mandatory before merge; security review depends on risk profile.
• 🔴 (13:33) Reviewers must run the full pnpm test, pnpm run typecheck, pnpm run lint, and pnpm run format:check commands and inspect exit codes; verify property tests across 10 runs; confirm the working tree is hash-verifiably unmodified afterward; and report each point as PASS, FAIL, CONCERN, or MUST-FIX with file:line evidence plus an overall MERGE or DO-NOT-MERGE verdict.
• 🔴 (13:34) packages/core/src/cgroup-memory.ts intentionally rereads /proc/self/cgroup and /proc/self/mountinfo on every discovery because membership can change during process lifetime; a worker must never be admitted from a stale cgroup directory or a cached discovery miss.
• 🟡 (13:34) resolveLinuxCgroupMemoryFiles() parses both cgroup v1 memory-controller and cgroup v2 memberships, prefers an explicit v1 memory controller on hybrid hosts, decodes procfs octal escapes, rejects non-absolute/NUL/.. paths, prefers the deepest direct mount root, and accepts a namespace-relative interpretation only when exactly one candidate is available.
• 🟡 (13:34) readLinuxCgroupMemoryHeadroomBytes() reads v2 memory.current against both memory.high and memory.max, or v1 memory.usage_in_bytes against memory.limit_in_bytes; it walks from the process cgroup to the mount point and returns the minimum finite max(0, limit - usage) headroom across boundaries.
• 🟡 (13:34) Cgroup memory discovery fails closed to undefined for malformed/unsafe paths, unreadable or malformed usage/limit data, ambiguous discovery, unsafe parent traversal, or other exceptions. The only tolerated missing-boundary case is the unresource-controlled cgroup v2 root where directory === mountPoint and root === "/".
• 🟡 (13:34) Assistant identified the main remaining correctness risk as the boundary among cgroup headroom, token-cap sizing, and worker growth; planned an end-to-end call-site trace, with any change kept narrow and accompanied by a focused counterexample before running the full suite.