Add local relayflow launcher and execute backlog F8b

[kjgbot]

The existing SDK could compile local hello, but this fresh checkout had no running daemon. The five legacy swarm YAMLs also fail the new compiler before execution. This adds a launcher that starts a local cell, submits through the journal protocol, captures the journal, and stops its daemon.

workflows/drive-local.yaml executed the actual BACKLOG F8b package: select the entry, rename the private retry validator to describe the authoring restriction, run existing tests, and capture the diff. The complete commands, failures, outputs, workflow dependency matrix and reproduction instructions are in ops/RUNTIME-STATUS.md. Scaffold commit 89f2f1d is the starting point for reproducing the one-time package.

Scope: deterministic local execution is proven without Cloud/Daytona/Relaycast. Agent worker wiring is exercised with a deterministic stream-only wrapper. This does not migrate the legacy autonomous drive/review/delivery loop, run a live model, implement workspace recovery, or change Cloud admission/authentication. A human owns merge.

Validation (full literal output committed under ops/runtime-evidence/):

$ node --test scripts/run-local-workflow.test.mjs
ℹ tests 4
ℹ pass 4
ℹ fail 0

The local flow's verification step runs:

npm_config_cache="$PWD/.relayflow/npm-cache" npm --userconfig /tmp/empty-npmrc run build --prefix packages/sdk
cd packages/sdk
node node_modules/vitest/vitest.mjs run tests/spec-parity.test.ts tests/cli.test.ts

 Test Files  2 passed (2)
      Tests  91 passed (91)

CARGO_HOME="$PWD/.cargo-home" CARGO_TARGET_DIR="$PWD/kernel/target" RUSTC="$HOME/.rustup/toolchains/stable-aarch64-apple-darwin/bin/rustc" "$HOME/.rustup/toolchains/stable-aarch64-apple-darwin/bin/cargo" test --manifest-path kernel/Cargo.toml --locked -p relayflowd --test crash_resume

test result: ok. 37 passed; 0 failed; 0 ignored; 0 measured; 0 filtered out; finished in 37.91s

Raw captured transcripts retain their original whitespace; the report records the unfiltered diff-check diagnostics and the source-only check separately. No review or merge signoff is claimed.

[coderabbitai]

Important

• 🔍 Trigger review

This repository does not receive automatic reviews because it has fewer than 10 stars.

⚙️ Run configuration

Configuration used: Organization UI

Review profile: CHILL

Plan: Team

Run ID: 51c7ea60-fc36-425c-ba79-8e5645c93625

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[cubic-dev-ai]

Review completed against the latest diff

<sub>Reply with feedback, questions, or to request a fix.

Re-trigger cubic</sub>

[kjgbot]

Review swarm: maintainability

No fresh transcript was produced for run 7845c989-5b9a-44dc-b613-9a10aa92622f (MISSING).

[kjgbot]

Review swarm: history

No fresh transcript was produced for run 7845c989-5b9a-44dc-b613-9a10aa92622f (MISSING).

[kjgbot]

Review swarm: structure

No fresh transcript was produced for run 7845c989-5b9a-44dc-b613-9a10aa92622f (MISSING).

[kjgbot]

🎯 review-swarm: FAILED (M:fail H:pass S:missing)

Lens transcripts posted as sibling comments above.

[kjgbot]

Spec review: original dc7a33a; fixes pushed on this PR branch at 521e07c. Leave open pending green CI and independent review.

The bounded local deterministic launcher serves RFC-0001 Gate 1 and §2 dogfooding. It talks to the existing journal protocol; decision #7 makes Relaycast optional projection, so absence of a cloud workspace is not a correctness defect. The F8b rename keeps the authoring restriction unchanged. This is not full Gate 7 acceptance, sealed-bundle deployment (decision #14), or proof of long-running agent execution.

Findings/fixes:

• ops/local-work-package.mjs:40: an interrupted write could truncate the target and make retries refuse it. Both the package record and source replacement now use a same-directory temporary file, flush before atomic rename, then directory fsync. The original byte/hash guard remains.
• scripts/run-local-workflow.test.mjs:10,20: fixtures/run directories leaked. Registered suite cleanup retains files until assertions finish, then removes the owned directories.
• ops/runtime-evidence/workflow-summary.txt:3: transformed JSON was presented alongside commands without identifying the transformation. Replaced it with fresh, full CLI output and exit codes for all seven workflows.
• The existing claim that drive-verify.txt:9 overstates test count is NOT reproduced. Running the exact source yields 63 CLI plus 28 parity tests, 91 total. Do not rewrite a real transcript based on a static miscount.

Captured commands/output (also committed under ops/runtime-evidence/spec-review-*.txt):

$ node --test scripts/run-local-workflow.test.mjs ops/local-work-package.test.mjs
✔ interrupted package write preserves the original and retry applies once (243.952833ms)
✔ local launcher journals deterministic effects and reads more than one journal page (601.889709ms)
✔ a failed command fails the run and prevents dependent effects (130.195167ms)
✔ the SDK worker completes an agent step through the local journal protocol (595.823584ms)
✔ missing daemon is refused before a data directory or run is created (62.254666ms)
ℹ tests 5
ℹ suites 0
ℹ pass 5
ℹ fail 0
ℹ cancelled 0
ℹ skipped 0
ℹ todo 0
ℹ duration_ms 1429.743458

exit_code=0

$ node node_modules/vitest/vitest.mjs run tests/spec-parity.test.ts tests/cli.test.ts

 RUN  v2.1.9 /Users/khaliqgant/AgentWorkforce/flows-runtime-0907-wt/packages/sdk

 ✓ tests/spec-parity.test.ts (28 tests) 153ms
 ✓ tests/cli.test.ts (63 tests) 2492ms
   ✓ flows check CLI > binds a checked relative wrapper to the flow directory for worker execution 474ms
   ✓ flows check CLI > uses the raw Claude adapter model flag instead of accepting auth status as model proof 306ms

 Test Files  2 passed (2)
      Tests  91 passed (91)
   Start at  23:44:42
   Duration  2.86s (transform 166ms, setup 0ms, collect 408ms, tests 2.64s, environment 0ms, prepare 76ms)


exit_code=0

The new kill-during-write test is failure injection, not a claim of full mutation verification. Existing kernel crash evidence in the PR is historical; I did not rerun the full kernel crash suite for this JS/package-write repair. The bounded launcher still documents missing worker heartbeat, workspace isolation/reset, and bare-LLM support; these are not proven by its wrapper wiring check.

CI at the new head still cannot review content. gh run view 34164420311 --log-failed captured:

review	Launch cloud swarm	2026-09-07T21:48:38.4926158Z Workflow prepare failed: 401 Unauthorized: Unauthorized

The CI credential owner must restore accepted authentication; independent reviewers must review 521e07c and close the outstanding threads based on the fixes and reproduction evidence. No merge or green-CI claim.

[github-actions]

Review swarm: FAILED

• maintainability: MISSING
• history: MISSING
• structure: MISSING

Cloud run: 7845c989-5b9a-44dc-b613-9a10aa92622f

[kjgbot]

Final disposition at 521e07c: LEFT OPEN.

The bounded local launcher/F8b changes pass this spec review after 521e07c; this is Gate 1/local protocol evidence, not full Gate 7 deployment or heartbeat/isolation proof. The remaining test-count thread has the literal rerun showing 63 CLI plus 28 parity tests; an independent reviewer must acknowledge that evidence and resolve the thread. Latest-head artifact and packed-consumer checks succeeded, but the review rerun failed with all three fresh lens transcripts MISSING after successful launch. Review infrastructure owner must restore the swarm and obtain real signoff at this head. Leave open; no merge.

Captured exact-head check query and output:

$ gh api repos/AgentWorkforce/flows/commits/521e07c175ecd09b058f490b470bf337709850af/check-runs --jq '[.check_runs[] | {name,head_sha,status,conclusion,details_url}]'
[{"conclusion":"failure","details_url":"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/AgentWorkforce/flows/actions/runs/34164420311/job/101873938274","head_sha":"521e07c175ecd09b058f490b470bf337709850af","name":"review","status":"completed"},{"conclusion":"success","details_url":"https://www.cubic.dev/pr/AgentWorkforce/flows/pull/231","head_sha":"521e07c175ecd09b058f490b470bf337709850af","name":"cubic · AI code reviewer","status":"completed"},{"conclusion":"success","details_url":"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/AgentWorkforce/flows/actions/runs/34164420328/job/101872524987","head_sha":"521e07c175ecd09b058f490b470bf337709850af","name":"linux-x64-artifact","status":"completed"},{"conclusion":"success","details_url":"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/AgentWorkforce/flows/actions/runs/34164420307/job/101872524549","head_sha":"521e07c175ecd09b058f490b470bf337709850af","name":"packed-consumer","status":"completed"}]

exit_code=0

Current swarm report: #231 (comment)

Review swarm: FAILED

• maintainability: MISSING
• history: MISSING
• structure: MISSING

Cloud run: 7845c989-5b9a-44dc-b613-9a10aa92622f