Slice 1: Job-as-product (JobTask, keyed run, observer, delegation)

[christopherkarani]

Slice 1: Job-as-product

• New public JobTask (name, agent, brief, expected-output contract) and JobProduct (results keyed by task name + merged summary).
• Job.run(_:tasks:merge:observer:) convenience reusing Workflow.MergeStrategy for the summary (one internal visibility widening: mergedOutput(from:) fileprivate -> internal; semantics unchanged).
• Optional AgentObserver threaded through JobSession.fanOut(_:observer:) into child agent.run calls (default nil, source-compatible).
• Manager-delegation recipe Job.delegate(_:manager:assignments:merge:observer:tokenLimit:prepare:) + JobAssignment: manager drafts briefs from notes, then one fan-out.
• One-fan-out gate kept; multi-round documented via shared store + decided-N.
• All new public types Sendable; TDD (failing test committed first).

Docs (additive, own sections): source DocC, README, new docs/guide/job-as-product.md (+ sidebar), overview.md row, front-facing-api.md subsection, api-catalog.md rows + refreshed counts (263/270 files, 2452 symbols). Example: Examples/JobAsProduct.

Test evidence

• SWARM_OMIT_INTEGRATION_TARGETS=1 swift test --filter JobAsProductTests — 9/9 pass (failed before impl: cannot find 'JobTask' in scope)
• --filter StoreBackedComposition — 28 pass; --filter WorkflowTests — 20 pass
• --filter DocumentationFreshnessTests — 35 pass
• ./scripts/ci/lean-build-test.sh — PASS
• swiftformat --lint on 5 touched files — 0 require formatting
• cd Examples/JobAsProduct && swift run JobAsProduct — builds and prints keyed results + merged summaries

[chatgpt-codex-connector]

You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard.