DEN-3945: consolidate all branches into main (2026-09-12 sweep)

[ORESoftware]

Consolidation of all branches into main — DEN-3945 fleet sweep (2026-09-12)

Linear: DEN-3945. Every local branch, worktree and uncommitted change under ~/codes was published without force; this PR merges every branch not already in main with merge commits only, conflicts resolved semantically per ORESoftware/my-ai/AGENTS.md. Durable sweep record: ORESoftware/den-3945-fleet-sweep.

Semantically resolved and merged

• origin/main (base advance) — Merge b25ed59, conflict-free, bringing the sweep up to origin/main (ec189ce..f0ba78e, 3 commits, PR ci: gate PRs on latest allowed Zed dependencies #19).
• automation/zed-latest-deps-gate-20260912 — Merge 3a7f649, conflict-free, changing exactly one file (+16/-4). Carries the single commit of this branch that never landed: e3f6bbc 'fix: support lockless packages in latest-deps gate'. PR ci: gate PRs on latest allowed Zed dependencies #19 merged 2026-09-14T01:24:09Z and took the branch's first two commits (1e036b1, 3fb848f, both on main); e3f6bbc was committed AFTERWARDS at 2026-09-14T23:19:01 and never proposed again, so this is a genuine outstanding follow-up, not a stale carrier - confirmed by git merge-base --is-ancestor e3f6bbc origin/main returning false. Semantically the fix stops the pre-push gate from failing repositories that legitimately have no tracked .zpkg.lock: it records whether the lock is tracked (git ls-files --error-unmatch), compares against HEAD only when it is, removes a generated lock on exit when it is not, and moves 'zed validate --require-lock' to AFTER 'zed install' so the lock exists by the time it is required. Note git diff between the branch and the sweep head also shows ~1200 deletions across .ores-lint/, analysis_options.yaml and eslint.config.mjs; those are files the BRANCH lacks because its merge base is older, not content the merge removes - the merge result touches one file only, which is why it was safe.

Not merged — already landed / superseded (evidence)

• macs-mbp/wip-20260912/main — c1-every-changed-path-landed, and now a strict ancestor of the sweep after origin/main was merged. Its sdk/package.json is the same stale blob 9a430ba5 described below. No PR was ever opened (pulls?head= empty at rc 0).
• macs-mbp/wip-20260912/main-170703 — STALE PRE-RXJS SNAPSHOT - merging it would REVERT landed work. classify reports one novel path, sdk/package.json, but the branch blob (9a430ba5) is an OLDER manifest than the sweep head's (b25e691): it deletes the './rxjs' exports subpath, the rxjs peerDependencies and peerDependenciesMeta entries, and the rxjs devDependency, and re-labels the package 'Zero runtime dependencies'. The RxJS boundary landed on main in bffd17e 'feat(DEN-3970): add a read-only RxJS lifecycle boundary' (2026-08-29), confirmed an ancestor of origin/main, and sdk/src/rxjs.mjs, sdk/src/rxjs.d.ts and sdk/test/rxjs.test.mjs are all present on the head while the branch tree has no rxjs file at all. The branch's merge base with the sweep is 2026-08-23T18:56:52Z, i.e. BEFORE the RxJS work, so the WIP snapshot simply captured a stale checkout. Merging would leave sdk/src/rxjs.mjs shipped but unreachable through the exports map, breaking any consumer importing @zed-pkg/sync/rxjs. No PR was ever opened (pulls?head= empty at rc 0).
• macs-mbp/wip-20260912/main-171532 — Identical case to macs-mbp/wip-20260912/main-170703: same single novel path sdk/package.json, carrying the byte-identical stale blob 9a430ba5 that strips the rxjs exports subpath, peerDependency and devDependency landed by bffd17e (DEN-3970). Merge base 2026-08-23T18:56:52Z predates that work. No PR was ever opened (pulls?head= empty at rc 0).
• wip/sync-preserve-20260911 — Same stale pre-RxJS manifest: its only novel path is sdk/package.json at the byte-identical blob 9a430ba5, which removes the './rxjs' export, the rxjs peerDependencies/peerDependenciesMeta and the rxjs devDependency that bffd17e (DEN-3970, on main) added. Merge base 2026-08-24T14:11:37Z predates that commit. Merging would revert the RxJS lifecycle boundary in the published package manifest. No PR was ever opened (pulls?head= empty at rc 0).

Not merged — the branch has its own open PR

• macs-mbp/integrate-20260912 — DEN-3945: consolidate macs-mbp local branches into main (2026-09-12 sweep) #20
• wip/land-uncommitted-20260906 — Land uncommitted work from the local checkout (5 paths) #14
• wip/land-uncommitted-20260905 — wip: land uncommitted working-tree changes (20260905) #11

Not merged — needs a human/product decision

• none

Local verification (exact head 3a7f6497b3d7)

ALL 26 origin branches were tested for containment, not just the 4 the claim listed. open-pr-heads rc=0 listing 3 open PRs (#20, #14, #11); none was merged by me. Clone made self-contained: repack -a -d, alternates removed, fsck --connectivity-only reports no missing or broken links. No case collision (the repo tracks lowercase agents.md only, and one PR-template spelling), so ordinary working-tree merges were safe. Stranded-work check: refs/local/chore/DEN-3956-tmp-temp-ignores is 1 commit ahead of origin/main but that commit is contained in the sweep already; the other 5 local refs are 0 ahead, so nothing exists only on local disk. RXJS REGRESSION CHECK, the main risk in this repo: three of the four WIP branches would have deleted the rxjs exports subpath and peer dependency from sdk/package.json, so after deciding not to merge them I verified the sweep head still matches origin/main exactly on that point - both have exports['./rxjs'] -> src/rxjs.mjs, peerDependencies.rxjs '>=7.8.2 <8', peerDependenciesMeta.rxjs.optional true and devDependencies.rxjs '^7.8.2'. The sweep's only sdk/package.json delta over main is additive: a lint:ores script, a prepublishOnly script and an oresLint metadata block. LOCKFILE CONSISTENCY: the sweep changes sdk/package.json but not sdk/package-lock.json, which is correct because the added keys are scripts and metadata rather than dependencies - proven by 'npm --prefix sdk ci' succeeding (added 6 packages, audited 7, 0 vulnerabilities), which fails outright when a lock and manifest disagree. Gates from .github/workflows run on the merged head: js-sdk job - npm ci OK, npm test 158 tests 154 pass 0 fail 4 skipped, npm run typecheck (tsc -p .) clean; rust-core job - cargo fmt --all --check clean, cargo clippy --all-targets -D warnings clean (rc 0), cargo test --all-targets all suites ok, 0 failed; ores-lint job - sh .ores-lint/lint.sh rc 0, 'ores-lint[require-send]: clean (13 files scanned)'; actionlint 0 findings on the merged head AND 0 on an extracted origin/main baseline. git diff --check origin/main HEAD clean. Not run: the wasm-parity job (needs wasm-bindgen and a from-source wasm rebuild) and the nix/dart/formal-methods/opto-sync resilience jobs - all unnecessary here by construction, since the sweep's entire delta over origin/main touches no Rust source, no sdk/src, and no protocol/ contract file (verified by git diff --name-only). Pushed after every merge commit; remote sweep head verified by ls-remote after each push and equals the local head. Disk 77 GB free; the cargo target dir under build-tmp, sdk/node_modules and the main baseline extraction were all deleted afterwards; the clone working tree is clean.

Contracts (TJSV)

not applicable, no TypeSpec/JSON Schema peer-authority pair exists in this repo: there are zero .tsp files (git ls-tree count 0), and the four JSON Schemas under protocol/ (change-event, write-policy, formal-app-lifecycle, formal-write-lifecycle) have no authored TypeSpec peer, so TJSV has no two authorities to compare and no workflow here invokes it. The sweep touches no file under protocol/ and no schema at all, so contract parity is unchanged by construction. No camelCase/snake_case question arose.

Merge policy

Draft until verified on the exact head. Where this org has no GitHub Actions minutes, zero-step jobs are admission refusals, not passes.

🤖 Generated with Claude Code

[linear-code]

DEN-3945