Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In `@test/e2e-scenario/live/double-onboard.test.ts`:
- Around line 338-341: The registryHas function currently substring-matches
REGISTRY_FILE contents which can false-positive; update registryHas to parse
REGISTRY_FILE as JSON and check membership against the registry's shape instead
of using includes. Specifically, inside registryHas parse
fs.readFileSync(REGISTRY_FILE, "utf8") to JSON and then test for the exact
sandboxName (e.g., if the registry is an object check
Object.prototype.hasOwnProperty.call(registry, sandboxName) or if it is an array
use registry.includes(sandboxName)); keep the fs.existsSync guard and handle
JSON parse errors gracefully.
- Around line 364-382: prerequisiteOrSkip currently only checks result.exitCode
but doesn't handle host.command throwing on spawn failures; wrap the await
host.command(...) call in a try/catch, capture any thrown error (from
host.command or spawn), build the same failure message (including
resultText-equivalent or error.message), and then follow the same logic: if
process.env.GITHUB_ACTIONS === "true" rethrow new Error(message) else call
skip(message); keep existing behavior for successful result and non-zero
exitCode paths (function: prerequisiteOrSkip, call: host.command).

---

Nitpick comments:
In `@test/e2e-scenario/live/double-onboard.test.ts`:
- Around line 349-361: The loop updates the local variable "last" but never uses
it, causing poor diagnostics on timeout; either surface it in the timeout path
(e.g., throw or return with a message including the "last" probe output) or mark
it intentionally unused by renaming "last" to "_last" to satisfy the repo rule;
update the polling block around the sandbox.openshell call (the variable "last"
and the final return false) accordingly so the test either fails with a
descriptive message containing the last probe output or compiles without
unused-variable lint warnings.

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In @.github/workflows/e2e-vitest-scenarios.yaml:
- Around line 264-265: The workflow only gates double-onboard-vitest, so when
inputs.scenarios=double-onboard the other free-standing jobs still run; update
the other free-standing jobs (openshell-version-pin-vitest,
onboard-negative-paths-vitest, and any similar standalone lanes) to use the same
selector-aware if condition (for example if: ${{ inputs.scenarios == '' ||
inputs.scenarios == 'double-onboard' }}), or alternatively add a single
selector-parsing job/step that emits a shared output (e.g., matrix or boolean
flags) and change each free-standing job to use that output in their if:
expressions so only matching scenarios run.

---

Outside diff comments:
In @.github/workflows/e2e-vitest-scenarios.yaml:
- Around line 47-64: The SCENARIOS handling currently only special-cases when
SCENARIOS equals "double-onboard" exactly; instead parse SCENARIOS as a
comma-separated list, remove the "double-onboard" element if present, and then
build args/matrix from the remaining IDs before invoking npx tsx
test/e2e-scenario/scenarios/run.ts; also set matrix="[]" when the list becomes
empty after removal. Update the downstream job conditional (the one that runs
the free-standing Vitest job, e.g., double-onboard-vitest) to check list
membership (presence of "double-onboard" in the parsed list) rather than
exact-string equality so the job runs when any dispatched input includes
"double-onboard". Ensure you reference SCENARIOS, args, matrix, run.ts, and the
free-standing double-onboard-vitest job when making these changes.