| 1 | /** |
| 2 | * #4131 WF-A3 — partial failure + synthesis. |
| 3 | * |
| 4 | * parallel() uses all-settled semantics: a failed slot becomes null; the run |
| 5 | * continues so a synthesizer can still produce an operator summary. The slot |
| 6 | * is null but not anonymous — `slots.errors` carries `{ index, kind, message }` |
| 7 | * for every drop, so the summary can name what was lost instead of guessing. |
| 8 | * |
| 9 | * Run: /workflow run docs/examples/dogfood-automatic/wf_a3_partial_failure_synthesis.workflow.js |
| 10 | * |
| 11 | * For pure VM proof without model spend, use workflow-js unit tests: |
| 12 | * cargo test -p codewhale-workflow-js --locked parallel_fan_out_maps_one_failure_to_null_slot |
| 13 | */ |
| 14 | export default async function () { |
| 15 | phase("Parallel scouts"); |
| 16 | const slots = await parallel([ |
| 17 | () => |
| 18 | task({ |
| 19 | description: "Healthy scout A", |
| 20 | label: "scout-a", |
| 21 | type: "explore", |
| 22 | prompt: "Return the string READY_A. Read-only.", |
| 23 | }), |
| 24 | // Give this child an intentionally tiny budget so it starts, then fails |
| 25 | // deterministically at the runtime boundary. A model refusal is still a |
| 26 | // successful transport-level completion, while response-schema failures |
| 27 | // intentionally abort the whole workflow so they remain loud. |
| 28 | () => |
| 29 | task({ |
| 30 | description: "Deliberately failing scout B", |
| 31 | label: "scout-b-fail", |
| 32 | type: "explore", |
| 33 | tokenBudget: 1, |
| 34 | prompt: |
| 35 | "Inspect Cargo.toml and return a detailed workspace summary. This child intentionally has a one-token budget so parallel() exercises a failed null slot.", |
| 36 | }), |
| 37 | () => |
| 38 | task({ |
| 39 | description: "Healthy scout C", |
| 40 | label: "scout-c", |
| 41 | type: "explore", |
| 42 | prompt: "Return the string READY_C. Read-only.", |
| 43 | }), |
| 44 | ]); |
| 45 | |
| 46 | phase("Synthesize"); |
| 47 | const surviving = (slots || []).filter((s) => s != null); |
| 48 | // The typed failure ledger: which slot died, and of what. |
| 49 | const dropped = (slots.errors || []).map( |
| 50 | (e) => `slot ${e.index} (${e.kind}): ${e.message}`, |
| 51 | ); |
| 52 | for (const line of dropped) { |
| 53 | log(`dropped: ${line}`); |
| 54 | } |
| 55 | const summary = await task({ |
| 56 | description: "Synthesize from surviving parallel slots", |
| 57 | label: "synthesizer", |
| 58 | type: "general", |
| 59 | prompt: [ |
| 60 | "Build one operator-facing summary from the surviving scout results.", |
| 61 | "Explicitly note which parallel slot failed, and why.", |
| 62 | `slot_count=${(slots || []).length} surviving=${surviving.length}`, |
| 63 | "dropped_slots:", |
| 64 | dropped.length ? dropped.join("\n") : "(none)", |
| 65 | "slots_json:", |
| 66 | JSON.stringify(slots), |
| 67 | ].join("\n"), |
| 68 | }); |
| 69 | |
| 70 | return { |
| 71 | scenario: "WF-A3", |
| 72 | slots, |
| 73 | surviving_count: surviving.length, |
| 74 | dropped_slots: dropped, |
| 75 | summary, |
| 76 | }; |
| 77 | } |
| 78 |