Bind the dispatch engine at onEnable

[os-warren]

Fixes #42

What

src/jobs/dispatch.job.ts (#43) exposed bindDispatchEngine(engine) but nothing called it, so duly_dispatch threw Job 'duly_dispatch' has no data engine … at every run and dispatched nothing — a scheduled job that appears fully configured and silently produces no tasks.

registerDulyActionHandlers in src/actions/register-handlers.ts is the one place an ObjectStack application is handed ctx.ql (via defineStack({ onEnable })), so it now also calls bindDispatchEngine(ql) beside the existing registerCatalogActionHandlers(ql) / registerTaskActionHandlers(ql) calls — one line, unchanged function shape.

HandlerRegistrationContext widened from { registerAction } to extends DispatchEngine (adds find / insert / update) so the type carries what the bind needs, single-sourced from dispatch.job.ts's own DispatchEngine interface rather than restated.

Per the issue, did not: rename registerDulyActionHandlers (the issue's "consider renaming" was optional; the dispatch prompt for this card said not to restructure the function), touch objectstack.config.ts, or change anything under src/jobs/.

Why the test is the deliverable

register-handlers.ts has no author-time gate — an unregistered/unbound handler renders as fully configured, is clickable/schedulable, and fails only at call time; pnpm validate is green either way. Every existing dispatch assertion (test/dispatch.test.ts) calls bindDispatchEngine(data) itself before touching dulyDispatch, which is right for testing the planner and the idempotency index — but it would keep passing even if this issue's one-line fix were reverted.

test/dispatch-wiring.test.ts boots the app the way a real host does instead: it merges objectstack.config.ts's default export with its onEnable named export before constructing AppPlugin — { ...stackConfig, onEnable } — exactly as @objectstack/cli's serve.ts does (measured in its dist/commands/serve.js, comment included in the test). test/task-actions.test.ts deliberately does not do this (passes the default export alone, and registers handlers by hand) — this is the first suite in the repo that does. It then calls dulyDispatch({ jobId: DISPATCH_JOB_NAME }) with no test-side bindDispatchEngine call and asserts a task is actually created against the real engine.

test/catalog-instantiate.test.ts's handler-wiring fake needed a matching update (no-op find/insert/update) to satisfy the widened interface — that suite is about the action-handler registry, not the dispatch engine, so the additions are inert.

Ablation (required by the dispatch prompt)

Removed the bindDispatchEngine(ql); line (plus its comment) from register-handlers.ts, confirmed on disk before measuring:

$ grep -c "bindDispatchEngine(ql);" src/actions/register-handlers.ts
0
$ git diff --stat -- src/actions/register-handlers.ts
 src/actions/register-handlers.ts | 7 +------
 1 file changed, 1 insertion(+), 6 deletions(-)

Then measured:

• pnpm validate → exit 0, green (no gate exists for this — as the issue predicted)
• pnpm test -- test/dispatch-wiring.test.ts → red, exactly:

  Error: Job 'duly_dispatch' has no data engine. The platform invokes a job handler with
  { jobId, data, bundle } and no engine handle, so the host must call
  bindDispatchEngine(ctx.ql) from defineStack({ onEnable }). No tasks were dispatched.
  

  (Test Files 1 failed | 8 passed (9), Tests 1 failed | 278 passed (279) — only the new suite fails)

Restored via a trap 'git checkout -- "$F"' EXIT INT TERM from the committed fix (commit made before ablating, per the "never ablate uncommitted work" rule), confirmed clean (git status --short empty, line back in place) before re-running gates.

Gates — all green at ee3640f

$ pnpm validate    # exit 0 — ✓ Validation passed
$ pnpm typecheck   # exit 0 — tsc --noEmit, no errors
$ pnpm test        # exit 0 — 9 test files, 279/279 passed
$ pnpm build       # exit 0 — ✓ Build complete, artifact + runtime bundle written

Known limit (unchanged from the issue, not this PR's to fix)

This closes the objectstack dev/config-boot path only. The objectstack build artifact path still cannot carry onEnable (a JSON artifact holds no functions), so an artifact-served boot never runs it and the binding never happens there — that's the second half of objectstack#14094, upstream, and applies identically to the action handlers already registered in this same function.

---

Generated by Claude Code