Distillation

Date: Sep 16, 2026

• 🔴 (23:41) User stated this is a Jujutsu repository because /home/byk/Code/opencode-lore-embedding-runtime/.jj/ exists; mutations must use jj, not git, and colocated Git tooling is permitted only for read-only operations or unsupported operations such as git submodule.
• 🔴 (23:41) User stated jj never fails on conflict: operations record conflicts in the resulting commit, after which jj st must be checked and conflicts resolved manually rather than with interactive jj resolve.
• 🔴 (23:41) User’s jj workflow requires: never use interactive flags; always pass -m "msg" when describing/committing; verify mutations with jj st and jj log; prefer stable change IDs over hexadecimal commit IDs; never rebase or describe immutable commits; and use jj undo or jj op restore <op-id> for recovery.
• 🔴 (23:41) User stated tests use a temporary SQLite database through packages/core/test/setup.ts, never the production database.
• 🔴 (23:41) User stated releases use Sentry Craft and versions or CHANGELOG.md must never be manually bumped or edited; release workflow commands are gh workflow run release.yml -f version=auto or an explicit version such as gh workflow run release.yml -f version=0.23.0.
• 🔴 [enforced-testing-discipline] (23:41) User requires failing-first deterministic regression tests for every defect: each test must fail on the base branch, pass with the fix, drive the real precondition, and fail if the tested guard is deleted. User stated, “Never trust the fix without seeing it fail.”
• 🔴 [enforced-review-workflow] (23:41) User’s review standard requires two distinct passes for non-trivial work: 1. adversarial/correctness review covering 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 leakage, protocol attacks, and privilege escalation.
• 🔴 [enforced-review-workflow] (23:41) User requires reviewers to run pnpm test, pnpm run typecheck, pnpm run lint, and pnpm run format:check; check exit codes; verify property tests across ×10 runs; hash-verify that the working tree remains unmodified; and report per-point PASS / FAIL / CONCERN / MUST-FIX verdicts with file:line evidence plus an overall MERGE / DO-NOT-MERGE verdict.
• 🔴 (23:41) User’s project review heuristics identify skip/early-return branches as the highest-risk surface; every such branch needs a test that makes it fire incorrectly and proves the operation is not skipped.
• 🔴 (23:41) User’s project requires fan-out registry additions to be covered by their contract batteries: SYNCED_TABLES by packages/core/test/sync-registry-contract.test.ts, MIGRATIONS by the db.test.ts schema-version assertion plus migration-specific tests, and AGENTS by agents.test.ts.
• 🔴 (23:41) User’s project treats frozen shared arrays as defensive protection for memoized results; mutation should fail loudly with TypeError in strict mode rather than silently corrupt shared cache state.
• 🔴 (23:41) User stated db().query monkeypatches are shadowed by the tracing Proxy; tests counting SQL statements must use log.registerSink({ withDbSpan }).
• 🟡 (23:42) Current working-copy candidate is Jujutsu change tmzmtlss, commit 8888d203b173, bookmark fix/embedding-runtime-admission, description fix(core): correct embedding runtime selection and admission, parent change pxrvwsvv / commit 2cbf4006 on main, with tree hash 134d3de918ab.
• 🟡 (23:42) Current candidate contains 20 changed files with exactly 1,929 insertions and 205 deletions; review-relevant changes include newly added packages/core/src/cgroup-memory.ts and packages/core/test/cgroup-memory.test.ts, plus modifications to packages/core/src/embedding/local.ts, packages/core/src/embedding/pool.ts, packages/core/src/embedding-cap.ts, packages/core/test/embedding-cap.test.ts, packages/core/test/embedding-pool-memory.test.ts, and packages/core/test/embedding-pool.test.ts.
• 🟡 (23:42) jj reported a pre-existing conflicted bookmark named fix/responses-private-references.
• 🔴 (23:42) User requires cgroup memory discovery to be re-read from /proc/self/cgroup and /proc/self/mountinfo each time so a worker is never admitted from a stale cgroup directory or a cached discovery miss.
• 🟡 (23:42) packages/core/src/cgroup-memory.ts adds Linux cgroup v1/v2 discovery and headroom calculation: parses /proc paths with octal escape decoding and traversal/NUL rejection; prefers an explicit v1 memory controller on hybrid hosts; matches cgroup mounts from mountinfo; supports unambiguous namespaced roots; reads v2 memory.current, memory.high, and memory.max or v1 memory.usage_in_bytes and memory.limit_in_bytes; computes the minimum finite ancestor headroom; and returns undefined on malformed, ambiguous, unreadable, or unsafe state.
• 🟡 (23:42) readLinuxCgroupMemoryHeadroomBytes() implements fail-closed behavior for unknown cgroup headroom: malformed numeric values, missing required files, unreadable descendant boundaries, unsafe parent traversal, discovery failure, and other exceptions return undefined; finite negative headroom is clamped to 0.
• 🟡 (23:42) In cgroup v2, the implementation permits the mount root itself to omit memory.current, memory.high, or memory.max only when directory === files.mountPoint and files.root === "/"; all descendant boundaries must remain readable to avoid subtracting sibling usage from the wrong scope.