self-bench 0.3.5 → 0.3.8

package/dist/temporal/activities/provenance.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"provenance.js","sourceRoot":"","sources":["../../../src/temporal/activities/provenance.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,OAAO,EAAE,MAAM,sBAAsB,CAAC;AAC/C,OAAO,EAAE,kBAAkB,EAAE,MAAM,oBAAoB,CAAC;AAGxD,OAAO,EAAE,kCAAkC,EAAE,oBAAoB,EAAE,MAAM,qBAAqB,CAAC;AAC/F,OAAO,EAAE,WAAW,EAAE,MAAM,4BAA4B,CAAC;AACzD,OAAO,EAAE,eAAe,EAAE,sBAAsB,EAAE,MAAM,cAAc,CAAC;AAEvE,MAAM,CAAC,KAAK,UAAU,oBAAoB,CACxC,KAAoB,EACpB,GAAe;IAEf,OAAO,MAAM,sBAAsB,CACjC,wCAAwC,EACxC,KAAK,EAAE,EAAE,MAAM,EAAE,EAAE,EAAE;QACnB,MAAM,CAAC,UAAU,EAAE,KAAK,CAAC,GAAG,MAAM,OAAO,CAAC,GAAG,CAAC,CAAC,KAAK,CAAC,GAAG,CAAC,GAAG,CAAC,UAAU,CAAC,EAAE,WAAW,EAAE,CAAC,CAAC,CAAC;QAC1F,MAAM,KAAK,GAAG,eAAe,CAAC,UAAU,CAAC,CAAC;QAC1C,MAAM,MAAM,GAAG,MAAM,kCAAkC,CAAC,GAAG,CAAC,UAAU,CAAC,GAAG,EAAE,KAAK,EAAE,MAAM,CAAC,CAAC;QAC3F,MAAM,QAAQ,GAAG,oBAAoB,CAAC,GAAG,CAAC,UAAU,CAAC,GAAG,EAAE,KAAK,EAAE,MAAM,CAAC,CAAC;QACzE,IAAI,QAAQ,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;YAC1B,MAAM,kBAAkB,CAAC,YAAY,CACnC,wEAAwE,EACxE,cAAc,CACf,CAAC;QACJ,CAAC;QACD,OAAO,MAAM,KAAK,CAAC,GAAG,CACpB,QAAQ,GAAG,CAAC,KAAK,sCAAsC,OAAO,CAAC,OAAO,EAAE,CAAC,IAAI,CAAC,OAAO,QAAQ,EAC7F,MAAM,CAAC,IAAI,CAAC,GAAG,QAAQ,CAAC,GAAG,CAAC,CAAC,OAAO,EAAE,EAAE,CAAC,IAAI,CAAC,SAAS,CAAC,OAAO,CAAC,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,IAAI,CAAC,EACjF,sBAAsB,CACvB,CAAC;IACJ,CAAC,CACF,CAAC;AACJ,CAAC"}

package/dist/temporal/activities/replay.d.ts

@@ -0,0 +1,20 @@
+import { z } from "zod";
+import type { ArtifactStore } from "../../artifacts.js";
+import { type Candidate, type ReplayMaterial, type ReplayRunRequest } from "../../contracts.js";
+import { provenanceMessageSchema } from "../../provenance.js";
+export interface ReplayServices {
+    /** Resolves the PR's completed commit the way discovery does (merge commit, else head). */
+    resolveCompletedCommit(repository: string, pullRequest: number): Promise<string>;
+    /** Current default-branch head, recorded as the run's repository commit. */
+    resolveRepositoryHead(repository: string): Promise<string>;
+}
+/**
+ * Rebuilds candidates from a source run's artifacts so previously failed candidates can be
+ * replayed through fresh authoring and verification without discovery.
+ */
+export declare function rebuildReplayCandidates(store: ArtifactStore, input: ReplayRunRequest, services?: ReplayServices): Promise<ReplayMaterial>;
+export declare function rebuildReplayCandidate(store: ArtifactStore, sourceRunId: string, candidateId: string, services: ReplayServices): Promise<{
+    candidate: Candidate;
+    provenanceMessage: z.infer<typeof provenanceMessageSchema>;
+}>;
+//# sourceMappingURL=replay.d.ts.map

package/dist/temporal/activities/replay.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"replay.d.ts","sourceRoot":"","sources":["../../../src/temporal/activities/replay.ts"],"names":[],"mappings":"AACA,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AACxB,OAAO,KAAK,EAAE,aAAa,EAAE,MAAM,oBAAoB,CAAC;AACxD,OAAO,EACL,KAAK,SAAS,EAGd,KAAK,cAAc,EACnB,KAAK,gBAAgB,EACtB,MAAM,oBAAoB,CAAC;AAG5B,OAAO,EAAE,uBAAuB,EAAE,MAAM,qBAAqB,CAAC;AAe9D,MAAM,WAAW,cAAc;IAC7B,2FAA2F;IAC3F,sBAAsB,CAAC,UAAU,EAAE,MAAM,EAAE,WAAW,EAAE,MAAM,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC;IACjF,4EAA4E;IAC5E,qBAAqB,CAAC,UAAU,EAAE,MAAM,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC;CAC5D;AAED;;;GAGG;AACH,wBAAsB,uBAAuB,CAC3C,KAAK,EAAE,aAAa,EACpB,KAAK,EAAE,gBAAgB,EACvB,QAAQ,GAAE,cAAsC,GAC/C,OAAO,CAAC,cAAc,CAAC,CA4BzB;AAED,wBAAsB,sBAAsB,CAC1C,KAAK,EAAE,aAAa,EACpB,WAAW,EAAE,MAAM,EACnB,WAAW,EAAE,MAAM,EACnB,QAAQ,EAAE,cAAc,GACvB,OAAO,CAAC;IAAE,SAAS,EAAE,SAAS,CAAC;IAAC,iBAAiB,EAAE,CAAC,CAAC,KAAK,CAAC,OAAO,uBAAuB,CAAC,CAAA;CAAE,CAAC,CAgD/F"}

package/dist/temporal/activities/replay.js

@@ -0,0 +1,135 @@
+import { ApplicationFailure } from "@temporalio/common";
+import { z } from "zod";
+import { candidateSchema, commitSchema, } from "../../contracts.js";
+import { githubRepository } from "../../github.js";
+import { runCommand } from "../../process.js";
+import { provenanceMessageSchema } from "../../provenance.js";
+import { githubToken } from "../../subscription-auth.js";
+import { safeHeartbeat } from "./runtime.js";
+const MAX_DISCOVERY_ATTEMPTS = 5;
+const storedProvenanceSchema = z.object({ source: provenanceMessageSchema });
+const discoveryReportSchema = z.object({ candidates: z.array(candidateSchema) });
+const storedDefinitionSchema = z.object({
+    difficulty: z.enum(["easy", "medium", "hard"]),
+    baseCommit: commitSchema,
+    sourcePr: z.number().int().positive(),
+    sourceUrl: z.string().url(),
+});
+/**
+ * Rebuilds candidates from a source run's artifacts so previously failed candidates can be
+ * replayed through fresh authoring and verification without discovery.
+ */
+export async function rebuildReplayCandidates(store, input, services = defaultReplayServices) {
+    const candidates = [];
+    const messages = [];
+    for (const candidateId of input.replay.candidateIds) {
+        safeHeartbeat(`rebuilding ${candidateId} from ${input.replay.sourceRunId}`);
+        const rebuilt = await rebuildReplayCandidate(store, input.replay.sourceRunId, candidateId, services);
+        candidates.push(rebuilt.candidate);
+        messages.push(JSON.stringify(rebuilt.provenanceMessage));
+    }
+    const first = candidates[0];
+    if (!first) {
+        throw ApplicationFailure.nonRetryable("replay has no candidates", "InvalidReplay");
+    }
+    const repositoryUrl = repositoryUrlFromPullRequest(first.sourceUrl);
+    const [provenance, commit] = await Promise.all([
+        store.put(`runs/${input.runId}/provenance/replay.jsonl`, Buffer.from(`${messages.join("\n")}\n`), "application/x-ndjson"),
+        services.resolveRepositoryHead(githubRepository(repositoryUrl)),
+    ]);
+    return { candidates, repository: { url: repositoryUrl, commit }, provenance };
+}
+export async function rebuildReplayCandidate(store, sourceRunId, candidateId, services) {
+    const provenanceKey = `runs/${sourceRunId}/provenance/${candidateId}.json`;
+    const provenanceBytes = await store.getByKey(provenanceKey);
+    if (!provenanceBytes) {
+        throw ApplicationFailure.nonRetryable(`replay candidate ${candidateId} has no provenance artifact at ${provenanceKey}`, "ReplayCandidateMissing");
+    }
+    const provenanceMessage = storedProvenanceSchema.parse(JSON.parse(Buffer.from(provenanceBytes).toString("utf8"))).source;
+    const provenance = await store.put(provenanceKey, provenanceBytes, "application/json");
+    const discovered = await findDiscoveredCandidate(store, sourceRunId, candidateId);
+    if (discovered) {
+        return {
+            candidate: candidateSchema.parse({
+                ...discovered,
+                request: provenanceMessage.content,
+                provenance,
+            }),
+            provenanceMessage,
+        };
+    }
+    const definition = await findDefinition(store, sourceRunId, candidateId);
+    if (!definition) {
+        throw ApplicationFailure.nonRetryable(`replay candidate ${candidateId} has neither a discovery report nor an authored definition in ${sourceRunId}`, "ReplayCandidateMissing");
+    }
+    const completedCommit = await services.resolveCompletedCommit(githubRepository(repositoryUrlFromPullRequest(definition.sourceUrl)), definition.sourcePr);
+    return {
+        candidate: candidateSchema.parse({
+            candidateId,
+            difficulty: definition.difficulty,
+            sourcePr: definition.sourcePr,
+            sourceUrl: definition.sourceUrl,
+            baseCommit: definition.baseCommit.toLowerCase(),
+            completedCommit: completedCommit.toLowerCase(),
+            request: provenanceMessage.content,
+            provenance,
+        }),
+        provenanceMessage,
+    };
+}
+async function findDiscoveredCandidate(store, sourceRunId, candidateId) {
+    const match = /^w(\d+)s(\d+)-/.exec(candidateId);
+    if (!match) {
+        return undefined;
+    }
+    for (let attempt = 1; attempt <= MAX_DISCOVERY_ATTEMPTS; attempt += 1) {
+        const key = `runs/${sourceRunId}/discovery/wave-${match[1]}/shard-${match[2]}/attempt-${attempt}/report.json`;
+        const bytes = await store.getByKey(key);
+        if (!bytes) {
+            continue;
+        }
+        const report = discoveryReportSchema.parse(JSON.parse(Buffer.from(bytes).toString("utf8")));
+        const found = report.candidates.find((candidate) => candidate.candidateId === candidateId);
+        if (found) {
+            return found;
+        }
+    }
+    return undefined;
+}
+async function findDefinition(store, sourceRunId, candidateId) {
+    const prefix = `runs/${sourceRunId}/authoring/${candidateId}`;
+    for (const key of [`${prefix}/definition.json`, `${prefix}/round-1/definition.json`]) {
+        const bytes = await store.getByKey(key);
+        if (bytes) {
+            return storedDefinitionSchema.parse(JSON.parse(Buffer.from(bytes).toString("utf8")));
+        }
+    }
+    return undefined;
+}
+function repositoryUrlFromPullRequest(sourceUrl) {
+    const match = /^https:\/\/github\.com\/([^/]+)\/([^/]+)\/pull\/\d+\/?$/i.exec(sourceUrl);
+    if (!match?.[1] || !match[2]) {
+        throw ApplicationFailure.nonRetryable(`invalid pull request URL ${sourceUrl}`, "InvalidReplay");
+    }
+    return `https://github.com/${match[1]}/${match[2]}.git`;
+}
+const defaultReplayServices = {
+    async resolveCompletedCommit(repository, pullRequest) {
+        const result = await runCommand("gh", ["pr", "view", String(pullRequest), "--repo", repository, "--json", "mergeCommit,headRefOid"], { env: await githubEnvironment() });
+        const parsed = JSON.parse(result.stdout);
+        const commit = parsed.mergeCommit?.oid ?? parsed.headRefOid;
+        if (!commit) {
+            throw new Error(`pull request ${repository}#${pullRequest} has no merge or head commit`);
+        }
+        return commit;
+    },
+    async resolveRepositoryHead(repository) {
+        const result = await runCommand("gh", ["api", `repos/${repository}/commits/HEAD`, "--jq", ".sha"], { env: await githubEnvironment() });
+        return commitSchema.parse(result.stdout.trim());
+    },
+};
+async function githubEnvironment() {
+    const token = await githubToken();
+    return token ? { ...process.env, GH_TOKEN: token } : process.env;
+}
+//# sourceMappingURL=replay.js.map

package/dist/temporal/activities/replay.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"replay.js","sourceRoot":"","sources":["../../../src/temporal/activities/replay.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,kBAAkB,EAAE,MAAM,oBAAoB,CAAC;AACxD,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AAExB,OAAO,EAEL,eAAe,EACf,YAAY,GAGb,MAAM,oBAAoB,CAAC;AAC5B,OAAO,EAAE,gBAAgB,EAAE,MAAM,iBAAiB,CAAC;AACnD,OAAO,EAAE,UAAU,EAAE,MAAM,kBAAkB,CAAC;AAC9C,OAAO,EAAE,uBAAuB,EAAE,MAAM,qBAAqB,CAAC;AAC9D,OAAO,EAAE,WAAW,EAAE,MAAM,4BAA4B,CAAC;AACzD,OAAO,EAAE,aAAa,EAAE,MAAM,cAAc,CAAC;AAE7C,MAAM,sBAAsB,GAAG,CAAC,CAAC;AAEjC,MAAM,sBAAsB,GAAG,CAAC,CAAC,MAAM,CAAC,EAAE,MAAM,EAAE,uBAAuB,EAAE,CAAC,CAAC;AAC7E,MAAM,qBAAqB,GAAG,CAAC,CAAC,MAAM,CAAC,EAAE,UAAU,EAAE,CAAC,CAAC,KAAK,CAAC,eAAe,CAAC,EAAE,CAAC,CAAC;AACjF,MAAM,sBAAsB,GAAG,CAAC,CAAC,MAAM,CAAC;IACtC,UAAU,EAAE,CAAC,CAAC,IAAI,CAAC,CAAC,MAAM,EAAE,QAAQ,EAAE,MAAM,CAAC,CAAC;IAC9C,UAAU,EAAE,YAAY;IACxB,QAAQ,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,EAAE,CAAC,QAAQ,EAAE;IACrC,SAAS,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,EAAE;CAC5B,CAAC,CAAC;AASH;;;GAGG;AACH,MAAM,CAAC,KAAK,UAAU,uBAAuB,CAC3C,KAAoB,EACpB,KAAuB,EACvB,WAA2B,qBAAqB;IAEhD,MAAM,UAAU,GAAgB,EAAE,CAAC;IACnC,MAAM,QAAQ,GAAa,EAAE,CAAC;IAC9B,KAAK,MAAM,WAAW,IAAI,KAAK,CAAC,MAAM,CAAC,YAAY,EAAE,CAAC;QACpD,aAAa,CAAC,cAAc,WAAW,SAAS,KAAK,CAAC,MAAM,CAAC,WAAW,EAAE,CAAC,CAAC;QAC5E,MAAM,OAAO,GAAG,MAAM,sBAAsB,CAC1C,KAAK,EACL,KAAK,CAAC,MAAM,CAAC,WAAW,EACxB,WAAW,EACX,QAAQ,CACT,CAAC;QACF,UAAU,CAAC,IAAI,CAAC,OAAO,CAAC,SAAS,CAAC,CAAC;QACnC,QAAQ,CAAC,IAAI,CAAC,IAAI,CAAC,SAAS,CAAC,OAAO,CAAC,iBAAiB,CAAC,CAAC,CAAC;IAC3D,CAAC;IACD,MAAM,KAAK,GAAG,UAAU,CAAC,CAAC,CAAC,CAAC;IAC5B,IAAI,CAAC,KAAK,EAAE,CAAC;QACX,MAAM,kBAAkB,CAAC,YAAY,CAAC,0BAA0B,EAAE,eAAe,CAAC,CAAC;IACrF,CAAC;IACD,MAAM,aAAa,GAAG,4BAA4B,CAAC,KAAK,CAAC,SAAS,CAAC,CAAC;IACpE,MAAM,CAAC,UAAU,EAAE,MAAM,CAAC,GAAG,MAAM,OAAO,CAAC,GAAG,CAAC;QAC7C,KAAK,CAAC,GAAG,CACP,QAAQ,KAAK,CAAC,KAAK,0BAA0B,EAC7C,MAAM,CAAC,IAAI,CAAC,GAAG,QAAQ,CAAC,IAAI,CAAC,IAAI,CAAC,IAAI,CAAC,EACvC,sBAAsB,CACvB;QACD,QAAQ,CAAC,qBAAqB,CAAC,gBAAgB,CAAC,aAAa,CAAC,CAAC;KAChE,CAAC,CAAC;IACH,OAAO,EAAE,UAAU,EAAE,UAAU,EAAE,EAAE,GAAG,EAAE,aAAa,EAAE,MAAM,EAAE,EAAE,UAAU,EAAE,CAAC;AAChF,CAAC;AAED,MAAM,CAAC,KAAK,UAAU,sBAAsB,CAC1C,KAAoB,EACpB,WAAmB,EACnB,WAAmB,EACnB,QAAwB;IAExB,MAAM,aAAa,GAAG,QAAQ,WAAW,eAAe,WAAW,OAAO,CAAC;IAC3E,MAAM,eAAe,GAAG,MAAM,KAAK,CAAC,QAAQ,CAAC,aAAa,CAAC,CAAC;IAC5D,IAAI,CAAC,eAAe,EAAE,CAAC;QACrB,MAAM,kBAAkB,CAAC,YAAY,CACnC,oBAAoB,WAAW,kCAAkC,aAAa,EAAE,EAChF,wBAAwB,CACzB,CAAC;IACJ,CAAC;IACD,MAAM,iBAAiB,GAAG,sBAAsB,CAAC,KAAK,CACpD,IAAI,CAAC,KAAK,CAAC,MAAM,CAAC,IAAI,CAAC,eAAe,CAAC,CAAC,QAAQ,CAAC,MAAM,CAAC,CAAC,CAC1D,CAAC,MAAM,CAAC;IACT,MAAM,UAAU,GAAG,MAAM,KAAK,CAAC,GAAG,CAAC,aAAa,EAAE,eAAe,EAAE,kBAAkB,CAAC,CAAC;IACvF,MAAM,UAAU,GAAG,MAAM,uBAAuB,CAAC,KAAK,EAAE,WAAW,EAAE,WAAW,CAAC,CAAC;IAClF,IAAI,UAAU,EAAE,CAAC;QACf,OAAO;YACL,SAAS,EAAE,eAAe,CAAC,KAAK,CAAC;gBAC/B,GAAG,UAAU;gBACb,OAAO,EAAE,iBAAiB,CAAC,OAAO;gBAClC,UAAU;aACX,CAAC;YACF,iBAAiB;SAClB,CAAC;IACJ,CAAC;IACD,MAAM,UAAU,GAAG,MAAM,cAAc,CAAC,KAAK,EAAE,WAAW,EAAE,WAAW,CAAC,CAAC;IACzE,IAAI,CAAC,UAAU,EAAE,CAAC;QAChB,MAAM,kBAAkB,CAAC,YAAY,CACnC,oBAAoB,WAAW,iEAAiE,WAAW,EAAE,EAC7G,wBAAwB,CACzB,CAAC;IACJ,CAAC;IACD,MAAM,eAAe,GAAG,MAAM,QAAQ,CAAC,sBAAsB,CAC3D,gBAAgB,CAAC,4BAA4B,CAAC,UAAU,CAAC,SAAS,CAAC,CAAC,EACpE,UAAU,CAAC,QAAQ,CACpB,CAAC;IACF,OAAO;QACL,SAAS,EAAE,eAAe,CAAC,KAAK,CAAC;YAC/B,WAAW;YACX,UAAU,EAAE,UAAU,CAAC,UAAU;YACjC,QAAQ,EAAE,UAAU,CAAC,QAAQ;YAC7B,SAAS,EAAE,UAAU,CAAC,SAAS;YAC/B,UAAU,EAAE,UAAU,CAAC,UAAU,CAAC,WAAW,EAAE;YAC/C,eAAe,EAAE,eAAe,CAAC,WAAW,EAAE;YAC9C,OAAO,EAAE,iBAAiB,CAAC,OAAO;YAClC,UAAU;SACX,CAAC;QACF,iBAAiB;KAClB,CAAC;AACJ,CAAC;AAED,KAAK,UAAU,uBAAuB,CACpC,KAAoB,EACpB,WAAmB,EACnB,WAAmB;IAEnB,MAAM,KAAK,GAAG,gBAAgB,CAAC,IAAI,CAAC,WAAW,CAAC,CAAC;IACjD,IAAI,CAAC,KAAK,EAAE,CAAC;QACX,OAAO,SAAS,CAAC;IACnB,CAAC;IACD,KAAK,IAAI,OAAO,GAAG,CAAC,EAAE,OAAO,IAAI,sBAAsB,EAAE,OAAO,IAAI,CAAC,EAAE,CAAC;QACtE,MAAM,GAAG,GAAG,QAAQ,WAAW,mBAAmB,KAAK,CAAC,CAAC,CAAC,UAAU,KAAK,CAAC,CAAC,CAAC,YAAY,OAAO,cAAc,CAAC;QAC9G,MAAM,KAAK,GAAG,MAAM,KAAK,CAAC,QAAQ,CAAC,GAAG,CAAC,CAAC;QACxC,IAAI,CAAC,KAAK,EAAE,CAAC;YACX,SAAS;QACX,CAAC;QACD,MAAM,MAAM,GAAG,qBAAqB,CAAC,KAAK,CAAC,IAAI,CAAC,KAAK,CAAC,MAAM,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,QAAQ,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC;QAC5F,MAAM,KAAK,GAAG,MAAM,CAAC,UAAU,CAAC,IAAI,CAAC,CAAC,SAAS,EAAE,EAAE,CAAC,SAAS,CAAC,WAAW,KAAK,WAAW,CAAC,CAAC;QAC3F,IAAI,KAAK,EAAE,CAAC;YACV,OAAO,KAAK,CAAC;QACf,CAAC;IACH,CAAC;IACD,OAAO,SAAS,CAAC;AACnB,CAAC;AAED,KAAK,UAAU,cAAc,CAC3B,KAAoB,EACpB,WAAmB,EACnB,WAAmB;IAEnB,MAAM,MAAM,GAAG,QAAQ,WAAW,cAAc,WAAW,EAAE,CAAC;IAC9D,KAAK,MAAM,GAAG,IAAI,CAAC,GAAG,MAAM,kBAAkB,EAAE,GAAG,MAAM,0BAA0B,CAAC,EAAE,CAAC;QACrF,MAAM,KAAK,GAAG,MAAM,KAAK,CAAC,QAAQ,CAAC,GAAG,CAAC,CAAC;QACxC,IAAI,KAAK,EAAE,CAAC;YACV,OAAO,sBAAsB,CAAC,KAAK,CAAC,IAAI,CAAC,KAAK,CAAC,MAAM,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,QAAQ,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC;QACvF,CAAC;IACH,CAAC;IACD,OAAO,SAAS,CAAC;AACnB,CAAC;AAED,SAAS,4BAA4B,CAAC,SAAiB;IACrD,MAAM,KAAK,GAAG,0DAA0D,CAAC,IAAI,CAAC,SAAS,CAAC,CAAC;IACzF,IAAI,CAAC,KAAK,EAAE,CAAC,CAAC,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,CAAC,EAAE,CAAC;QAC7B,MAAM,kBAAkB,CAAC,YAAY,CAAC,4BAA4B,SAAS,EAAE,EAAE,eAAe,CAAC,CAAC;IAClG,CAAC;IACD,OAAO,sBAAsB,KAAK,CAAC,CAAC,CAAC,IAAI,KAAK,CAAC,CAAC,CAAC,MAAM,CAAC;AAC1D,CAAC;AAED,MAAM,qBAAqB,GAAmB;IAC5C,KAAK,CAAC,sBAAsB,CAAC,UAAU,EAAE,WAAW;QAClD,MAAM,MAAM,GAAG,MAAM,UAAU,CAC7B,IAAI,EACJ,CAAC,IAAI,EAAE,MAAM,EAAE,MAAM,CAAC,WAAW,CAAC,EAAE,QAAQ,EAAE,UAAU,EAAE,QAAQ,EAAE,wBAAwB,CAAC,EAC7F,EAAE,GAAG,EAAE,MAAM,iBAAiB,EAAE,EAAE,CACnC,CAAC;QACF,MAAM,MAAM,GAAG,IAAI,CAAC,KAAK,CAAC,MAAM,CAAC,MAAM,CAGtC,CAAC;QACF,MAAM,MAAM,GAAG,MAAM,CAAC,WAAW,EAAE,GAAG,IAAI,MAAM,CAAC,UAAU,CAAC;QAC5D,IAAI,CAAC,MAAM,EAAE,CAAC;YACZ,MAAM,IAAI,KAAK,CAAC,gBAAgB,UAAU,IAAI,WAAW,8BAA8B,CAAC,CAAC;QAC3F,CAAC;QACD,OAAO,MAAM,CAAC;IAChB,CAAC;IACD,KAAK,CAAC,qBAAqB,CAAC,UAAU;QACpC,MAAM,MAAM,GAAG,MAAM,UAAU,CAC7B,IAAI,EACJ,CAAC,KAAK,EAAE,SAAS,UAAU,eAAe,EAAE,MAAM,EAAE,MAAM,CAAC,EAC3D,EAAE,GAAG,EAAE,MAAM,iBAAiB,EAAE,EAAE,CACnC,CAAC;QACF,OAAO,YAAY,CAAC,KAAK,CAAC,MAAM,CAAC,MAAM,CAAC,IAAI,EAAE,CAAC,CAAC;IAClD,CAAC;CACF,CAAC;AAEF,KAAK,UAAU,iBAAiB;IAC9B,MAAM,KAAK,GAAG,MAAM,WAAW,EAAE,CAAC;IAClC,OAAO,KAAK,CAAC,CAAC,CAAC,EAAE,GAAG,OAAO,CAAC,GAAG,EAAE,QAAQ,EAAE,KAAK,EAAE,CAAC,CAAC,CAAC,OAAO,CAAC,GAAG,CAAC;AACnE,CAAC"}

package/dist/temporal/activities/round-outcome.d.ts

@@ -0,0 +1,63 @@
+import type { ArtifactStore } from "../../artifacts.js";
+/** Tool calls that end an agent session with a deliverable the worker must be able to collect. */
+export declare const TERMINAL_TOOLS: readonly ["submit_task", "submit_fix", "accept_task", "submit_suggestions", "reject_task"];
+/**
+ * The round wrapper's own exit status, written by its EXIT trap. Sandbox providers can lose a
+ * long-running command's stream after the script has finished (E2B reports a gRPC "terminated"
+ * or a stray exit code), so this file, not the provider's code, is what decides the round.
+ */
+export declare const WRAPPER_STATUS_PATH = "/work/wrapper-status";
+export interface SandboxRoundResult {
+    readonly sandboxId: string;
+    readonly exitCode: number;
+    /** The exit code the sandbox provider reported, when the wrapper status file overrode it. */
+    readonly providerExitCode?: number | undefined;
+    readonly stdout: string;
+    readonly stderr: string;
+    readonly outputs: Readonly<Record<string, Uint8Array>>;
+}
+/** The wrapper's recorded exit status, when the sandbox delivered the status file. */
+export declare function wrapperStatusFrom(outputs: Readonly<Record<string, Uint8Array>>): number | undefined;
+/**
+ * Replaces the provider-reported exit code with the wrapper's own status when the sandbox
+ * delivered it. A hard timeout is the provider's verdict and stays authoritative: the wrapper's
+ * trap may have recorded the kill signal instead.
+ */
+export declare function reconcileWrapperStatus<T extends SandboxRoundResult>(result: T): T;
+/** A delivered submission or verdict could not be collected; retrying the activity is correct. */
+export declare class SandboxOutputError extends Error {
+    constructor(message: string);
+}
+export type RoundVerdict = {
+    readonly kind: "ok";
+} | {
+    readonly kind: "rejected";
+    readonly reason: string;
+} | {
+    readonly kind: "infrastructure";
+    readonly reason: string;
+};
+export interface RoundClassification {
+    readonly round: number;
+    readonly exitCode: number;
+    /** pi's own exit code as printed by the wrapper, when it differs from the command's. */
+    readonly piExitCode?: number | undefined;
+    /** Declared outputs that were not collected. */
+    readonly missing: readonly string[];
+    readonly sessionCollected: boolean;
+    readonly toolCalls: readonly string[];
+    readonly finalMessage?: string | undefined;
+    /** The provider error that ended the session, if pi's last assistant turn failed. */
+    readonly providerError?: string | undefined;
+}
+/**
+ * Why a round did not deliver, phrased so a human can act on it. A delivered terminal tool call
+ * whose output is missing, and a sandbox that died before any terminal call, are infrastructure
+ * problems for Temporal to retry, never candidate rejections.
+ */
+export declare function classifyRound(input: RoundClassification): RoundVerdict;
+/** `[selfbench] pi exited with N` as printed by the round wrapper. */
+export declare function piExitCodeFrom(stdout: string): number | undefined;
+/** Archives exit code, collected outputs with sizes, and output tails next to the round. */
+export declare function archiveSandboxResult(store: ArtifactStore, key: string, result: SandboxRoundResult, declaredOutputs: readonly string[]): Promise<string[]>;
+//# sourceMappingURL=round-outcome.d.ts.map

package/dist/temporal/activities/round-outcome.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"round-outcome.d.ts","sourceRoot":"","sources":["../../../src/temporal/activities/round-outcome.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,aAAa,EAAE,MAAM,oBAAoB,CAAC;AAGxD,kGAAkG;AAClG,eAAO,MAAM,cAAc,4FAMjB,CAAC;AAEX;;;;GAIG;AACH,eAAO,MAAM,mBAAmB,yBAAyB,CAAC;AAE1D,MAAM,WAAW,kBAAkB;IACjC,QAAQ,CAAC,SAAS,EAAE,MAAM,CAAC;IAC3B,QAAQ,CAAC,QAAQ,EAAE,MAAM,CAAC;IAC1B,6FAA6F;IAC7F,QAAQ,CAAC,gBAAgB,CAAC,EAAE,MAAM,GAAG,SAAS,CAAC;IAC/C,QAAQ,CAAC,MAAM,EAAE,MAAM,CAAC;IACxB,QAAQ,CAAC,MAAM,EAAE,MAAM,CAAC;IACxB,QAAQ,CAAC,OAAO,EAAE,QAAQ,CAAC,MAAM,CAAC,MAAM,EAAE,UAAU,CAAC,CAAC,CAAC;CACxD;AAED,sFAAsF;AACtF,wBAAgB,iBAAiB,CAC/B,OAAO,EAAE,QAAQ,CAAC,MAAM,CAAC,MAAM,EAAE,UAAU,CAAC,CAAC,GAC5C,MAAM,GAAG,SAAS,CAOpB;AAED;;;;GAIG;AACH,wBAAgB,sBAAsB,CAAC,CAAC,SAAS,kBAAkB,EAAE,MAAM,EAAE,CAAC,GAAG,CAAC,CASjF;AAED,kGAAkG;AAClG,qBAAa,kBAAmB,SAAQ,KAAK;gBAC/B,OAAO,EAAE,MAAM;CAI5B;AAED,MAAM,MAAM,YAAY,GACpB;IAAE,QAAQ,CAAC,IAAI,EAAE,IAAI,CAAA;CAAE,GACvB;IAAE,QAAQ,CAAC,IAAI,EAAE,UAAU,CAAC;IAAC,QAAQ,CAAC,MAAM,EAAE,MAAM,CAAA;CAAE,GACtD;IAAE,QAAQ,CAAC,IAAI,EAAE,gBAAgB,CAAC;IAAC,QAAQ,CAAC,MAAM,EAAE,MAAM,CAAA;CAAE,CAAC;AAEjE,MAAM,WAAW,mBAAmB;IAClC,QAAQ,CAAC,KAAK,EAAE,MAAM,CAAC;IACvB,QAAQ,CAAC,QAAQ,EAAE,MAAM,CAAC;IAC1B,wFAAwF;IACxF,QAAQ,CAAC,UAAU,CAAC,EAAE,MAAM,GAAG,SAAS,CAAC;IACzC,gDAAgD;IAChD,QAAQ,CAAC,OAAO,EAAE,SAAS,MAAM,EAAE,CAAC;IACpC,QAAQ,CAAC,gBAAgB,EAAE,OAAO,CAAC;IACnC,QAAQ,CAAC,SAAS,EAAE,SAAS,MAAM,EAAE,CAAC;IACtC,QAAQ,CAAC,YAAY,CAAC,EAAE,MAAM,GAAG,SAAS,CAAC;IAC3C,qFAAqF;IACrF,QAAQ,CAAC,aAAa,CAAC,EAAE,MAAM,GAAG,SAAS,CAAC;CAC7C;AAED;;;;GAIG;AACH,wBAAgB,aAAa,CAAC,KAAK,EAAE,mBAAmB,GAAG,YAAY,CAiDtE;AAED,sEAAsE;AACtE,wBAAgB,cAAc,CAAC,MAAM,EAAE,MAAM,GAAG,MAAM,GAAG,SAAS,CAGjE;AAED,4FAA4F;AAC5F,wBAAsB,oBAAoB,CACxC,KAAK,EAAE,aAAa,EACpB,GAAG,EAAE,MAAM,EACX,MAAM,EAAE,kBAAkB,EAC1B,eAAe,EAAE,SAAS,MAAM,EAAE,GACjC,OAAO,CAAC,MAAM,EAAE,CAAC,CAkBnB"}

package/dist/temporal/activities/round-outcome.js

@@ -0,0 +1,127 @@
+import { boundedTail } from "./harbor.js";
+/** Tool calls that end an agent session with a deliverable the worker must be able to collect. */
+export const TERMINAL_TOOLS = [
+    "submit_task",
+    "submit_fix",
+    "accept_task",
+    "submit_suggestions",
+    "reject_task",
+];
+const HARD_TIMEOUT_EXIT_CODE = 124;
+/**
+ * The round wrapper's own exit status, written by its EXIT trap. Sandbox providers can lose a
+ * long-running command's stream after the script has finished (E2B reports a gRPC "terminated"
+ * or a stray exit code), so this file, not the provider's code, is what decides the round.
+ */
+export const WRAPPER_STATUS_PATH = "/work/wrapper-status";
+/** The wrapper's recorded exit status, when the sandbox delivered the status file. */
+export function wrapperStatusFrom(outputs) {
+    const bytes = outputs[WRAPPER_STATUS_PATH];
+    if (!bytes) {
+        return undefined;
+    }
+    const text = Buffer.from(bytes).toString("utf8").trim();
+    return /^\d{1,3}$/.test(text) ? Number(text) : undefined;
+}
+/**
+ * Replaces the provider-reported exit code with the wrapper's own status when the sandbox
+ * delivered it. A hard timeout is the provider's verdict and stays authoritative: the wrapper's
+ * trap may have recorded the kill signal instead.
+ */
+export function reconcileWrapperStatus(result) {
+    const status = wrapperStatusFrom(result.outputs);
+    if (status === undefined || result.exitCode === HARD_TIMEOUT_EXIT_CODE) {
+        return result;
+    }
+    if (status === result.exitCode) {
+        return result;
+    }
+    return { ...result, exitCode: status, providerExitCode: result.exitCode };
+}
+/** A delivered submission or verdict could not be collected; retrying the activity is correct. */
+export class SandboxOutputError extends Error {
+    constructor(message) {
+        super(message);
+        this.name = "SandboxOutputError";
+    }
+}
+/**
+ * Why a round did not deliver, phrased so a human can act on it. A delivered terminal tool call
+ * whose output is missing, and a sandbox that died before any terminal call, are infrastructure
+ * problems for Temporal to retry, never candidate rejections.
+ */
+export function classifyRound(input) {
+    const terminal = [...input.toolCalls]
+        .reverse()
+        .find((name) => TERMINAL_TOOLS.includes(name));
+    const pi = input.piExitCode ?? input.exitCode;
+    const explanation = input.finalMessage
+        ? `; agent said: ${input.finalMessage.slice(0, 1_000)}`
+        : "";
+    if (input.providerError && !terminal) {
+        return {
+            kind: "infrastructure",
+            reason: `round ${input.round}: the model provider failed mid-session (${input.providerError.slice(0, 200)}) before a terminal tool call`,
+        };
+    }
+    if (input.exitCode === 0 && input.missing.length === 0 && input.sessionCollected) {
+        return { kind: "ok" };
+    }
+    if (input.exitCode === 0) {
+        const what = input.missing.length > 0
+            ? `output ${input.missing.join(", ")} missing`
+            : "session output missing";
+        return {
+            kind: "infrastructure",
+            reason: `round ${input.round}: ${what} (pi exit 0${terminal ? ` after ${terminal}` : ""}); the sandbox delivered but collection failed`,
+        };
+    }
+    if (terminal) {
+        return {
+            kind: "infrastructure",
+            reason: `round ${input.round}: wrapper exited with code ${input.exitCode} after ${terminal} was recorded`,
+        };
+    }
+    if (input.exitCode === HARD_TIMEOUT_EXIT_CODE || !input.sessionCollected) {
+        return {
+            kind: "infrastructure",
+            reason: `round ${input.round}: sandbox died mid-round (exit ${input.exitCode}${input.sessionCollected ? "" : ", no session collected"}) before a terminal tool call`,
+        };
+    }
+    if (pi !== 0) {
+        return {
+            kind: "rejected",
+            reason: `round ${input.round}: pi exited with code ${pi}${explanation}`,
+        };
+    }
+    return {
+        kind: "rejected",
+        reason: `round ${input.round}: pi exited 0 without a terminal tool call and the wrapper exited with code ${input.exitCode}${explanation}`,
+    };
+}
+/** `[selfbench] pi exited with N` as printed by the round wrapper. */
+export function piExitCodeFrom(stdout) {
+    const match = /\[selfbench\] pi exited with (\d+)/.exec(stdout);
+    return match?.[1] === undefined ? undefined : Number(match[1]);
+}
+/** Archives exit code, collected outputs with sizes, and output tails next to the round. */
+export async function archiveSandboxResult(store, key, result, declaredOutputs) {
+    const missing = declaredOutputs.filter((path) => result.outputs[path] === undefined);
+    const record = {
+        sandboxId: result.sandboxId,
+        exitCode: result.exitCode,
+        providerExitCode: result.providerExitCode ?? null,
+        wrapperStatus: wrapperStatusFrom(result.outputs) ?? null,
+        piExitCode: piExitCodeFrom(result.stdout) ?? null,
+        outputs: declaredOutputs.map((path) => ({
+            path,
+            collected: result.outputs[path] !== undefined,
+            sizeBytes: result.outputs[path]?.length ?? 0,
+        })),
+        stdoutTail: boundedTail(result.stdout, 4_000),
+        stderrTail: boundedTail(result.stderr, 4_000),
+    };
+    await store.put(key, Buffer.from(`${JSON.stringify(record, null, 2)}\n`), "application/json");
+    return missing;
+}
+//# sourceMappingURL=round-outcome.js.map

package/dist/temporal/activities/round-outcome.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"round-outcome.js","sourceRoot":"","sources":["../../../src/temporal/activities/round-outcome.ts"],"names":[],"mappings":"AACA,OAAO,EAAE,WAAW,EAAE,MAAM,aAAa,CAAC;AAE1C,kGAAkG;AAClG,MAAM,CAAC,MAAM,cAAc,GAAG;IAC5B,aAAa;IACb,YAAY;IACZ,aAAa;IACb,oBAAoB;IACpB,aAAa;CACL,CAAC;AACX,MAAM,sBAAsB,GAAG,GAAG,CAAC;AACnC;;;;GAIG;AACH,MAAM,CAAC,MAAM,mBAAmB,GAAG,sBAAsB,CAAC;AAY1D,sFAAsF;AACtF,MAAM,UAAU,iBAAiB,CAC/B,OAA6C;IAE7C,MAAM,KAAK,GAAG,OAAO,CAAC,mBAAmB,CAAC,CAAC;IAC3C,IAAI,CAAC,KAAK,EAAE,CAAC;QACX,OAAO,SAAS,CAAC;IACnB,CAAC;IACD,MAAM,IAAI,GAAG,MAAM,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,QAAQ,CAAC,MAAM,CAAC,CAAC,IAAI,EAAE,CAAC;IACxD,OAAO,WAAW,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC;AAC3D,CAAC;AAED;;;;GAIG;AACH,MAAM,UAAU,sBAAsB,CAA+B,MAAS;IAC5E,MAAM,MAAM,GAAG,iBAAiB,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC;IACjD,IAAI,MAAM,KAAK,SAAS,IAAI,MAAM,CAAC,QAAQ,KAAK,sBAAsB,EAAE,CAAC;QACvE,OAAO,MAAM,CAAC;IAChB,CAAC;IACD,IAAI,MAAM,KAAK,MAAM,CAAC,QAAQ,EAAE,CAAC;QAC/B,OAAO,MAAM,CAAC;IAChB,CAAC;IACD,OAAO,EAAE,GAAG,MAAM,EAAE,QAAQ,EAAE,MAAM,EAAE,gBAAgB,EAAE,MAAM,CAAC,QAAQ,EAAE,CAAC;AAC5E,CAAC;AAED,kGAAkG;AAClG,MAAM,OAAO,kBAAmB,SAAQ,KAAK;IAC3C,YAAY,OAAe;QACzB,KAAK,CAAC,OAAO,CAAC,CAAC;QACf,IAAI,CAAC,IAAI,GAAG,oBAAoB,CAAC;IACnC,CAAC;CACF;AAqBD;;;;GAIG;AACH,MAAM,UAAU,aAAa,CAAC,KAA0B;IACtD,MAAM,QAAQ,GAAG,CAAC,GAAG,KAAK,CAAC,SAAS,CAAC;SAClC,OAAO,EAAE;SACT,IAAI,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,cAAc,CAAC,QAAQ,CAAC,IAAa,CAAC,CAAC,CAAC;IAC1D,MAAM,EAAE,GAAG,KAAK,CAAC,UAAU,IAAI,KAAK,CAAC,QAAQ,CAAC;IAC9C,MAAM,WAAW,GAAG,KAAK,CAAC,YAAY;QACpC,CAAC,CAAC,iBAAiB,KAAK,CAAC,YAAY,CAAC,KAAK,CAAC,CAAC,EAAE,KAAK,CAAC,EAAE;QACvD,CAAC,CAAC,EAAE,CAAC;IACP,IAAI,KAAK,CAAC,aAAa,IAAI,CAAC,QAAQ,EAAE,CAAC;QACrC,OAAO;YACL,IAAI,EAAE,gBAAgB;YACtB,MAAM,EAAE,SAAS,KAAK,CAAC,KAAK,4CAA4C,KAAK,CAAC,aAAa,CAAC,KAAK,CAAC,CAAC,EAAE,GAAG,CAAC,+BAA+B;SACzI,CAAC;IACJ,CAAC;IACD,IAAI,KAAK,CAAC,QAAQ,KAAK,CAAC,IAAI,KAAK,CAAC,OAAO,CAAC,MAAM,KAAK,CAAC,IAAI,KAAK,CAAC,gBAAgB,EAAE,CAAC;QACjF,OAAO,EAAE,IAAI,EAAE,IAAI,EAAE,CAAC;IACxB,CAAC;IACD,IAAI,KAAK,CAAC,QAAQ,KAAK,CAAC,EAAE,CAAC;QACzB,MAAM,IAAI,GACR,KAAK,CAAC,OAAO,CAAC,MAAM,GAAG,CAAC;YACtB,CAAC,CAAC,UAAU,KAAK,CAAC,OAAO,CAAC,IAAI,CAAC,IAAI,CAAC,UAAU;YAC9C,CAAC,CAAC,wBAAwB,CAAC;QAC/B,OAAO;YACL,IAAI,EAAE,gBAAgB;YACtB,MAAM,EAAE,SAAS,KAAK,CAAC,KAAK,KAAK,IAAI,cAAc,QAAQ,CAAC,CAAC,CAAC,UAAU,QAAQ,EAAE,CAAC,CAAC,CAAC,EAAE,gDAAgD;SACxI,CAAC;IACJ,CAAC;IACD,IAAI,QAAQ,EAAE,CAAC;QACb,OAAO;YACL,IAAI,EAAE,gBAAgB;YACtB,MAAM,EAAE,SAAS,KAAK,CAAC,KAAK,8BAA8B,KAAK,CAAC,QAAQ,UAAU,QAAQ,eAAe;SAC1G,CAAC;IACJ,CAAC;IACD,IAAI,KAAK,CAAC,QAAQ,KAAK,sBAAsB,IAAI,CAAC,KAAK,CAAC,gBAAgB,EAAE,CAAC;QACzE,OAAO;YACL,IAAI,EAAE,gBAAgB;YACtB,MAAM,EAAE,SAAS,KAAK,CAAC,KAAK,kCAAkC,KAAK,CAAC,QAAQ,GAAG,KAAK,CAAC,gBAAgB,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,wBAAwB,+BAA+B;SACrK,CAAC;IACJ,CAAC;IACD,IAAI,EAAE,KAAK,CAAC,EAAE,CAAC;QACb,OAAO;YACL,IAAI,EAAE,UAAU;YAChB,MAAM,EAAE,SAAS,KAAK,CAAC,KAAK,yBAAyB,EAAE,GAAG,WAAW,EAAE;SACxE,CAAC;IACJ,CAAC;IACD,OAAO;QACL,IAAI,EAAE,UAAU;QAChB,MAAM,EAAE,SAAS,KAAK,CAAC,KAAK,+EAA+E,KAAK,CAAC,QAAQ,GAAG,WAAW,EAAE;KAC1I,CAAC;AACJ,CAAC;AAED,sEAAsE;AACtE,MAAM,UAAU,cAAc,CAAC,MAAc;IAC3C,MAAM,KAAK,GAAG,oCAAoC,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;IAChE,OAAO,KAAK,EAAE,CAAC,CAAC,CAAC,KAAK,SAAS,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC;AACjE,CAAC;AAED,4FAA4F;AAC5F,MAAM,CAAC,KAAK,UAAU,oBAAoB,CACxC,KAAoB,EACpB,GAAW,EACX,MAA0B,EAC1B,eAAkC;IAElC,MAAM,OAAO,GAAG,eAAe,CAAC,MAAM,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,MAAM,CAAC,OAAO,CAAC,IAAI,CAAC,KAAK,SAAS,CAAC,CAAC;IACrF,MAAM,MAAM,GAAG;QACb,SAAS,EAAE,MAAM,CAAC,SAAS;QAC3B,QAAQ,EAAE,MAAM,CAAC,QAAQ;QACzB,gBAAgB,EAAE,MAAM,CAAC,gBAAgB,IAAI,IAAI;QACjD,aAAa,EAAE,iBAAiB,CAAC,MAAM,CAAC,OAAO,CAAC,IAAI,IAAI;QACxD,UAAU,EAAE,cAAc,CAAC,MAAM,CAAC,MAAM,CAAC,IAAI,IAAI;QACjD,OAAO,EAAE,eAAe,CAAC,GAAG,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,CAAC;YACtC,IAAI;YACJ,SAAS,EAAE,MAAM,CAAC,OAAO,CAAC,IAAI,CAAC,KAAK,SAAS;YAC7C,SAAS,EAAE,MAAM,CAAC,OAAO,CAAC,IAAI,CAAC,EAAE,MAAM,IAAI,CAAC;SAC7C,CAAC,CAAC;QACH,UAAU,EAAE,WAAW,CAAC,MAAM,CAAC,MAAM,EAAE,KAAK,CAAC;QAC7C,UAAU,EAAE,WAAW,CAAC,MAAM,CAAC,MAAM,EAAE,KAAK,CAAC;KAC9C,CAAC;IACF,MAAM,KAAK,CAAC,GAAG,CAAC,GAAG,EAAE,MAAM,CAAC,IAAI,CAAC,GAAG,IAAI,CAAC,SAAS,CAAC,MAAM,EAAE,IAAI,EAAE,CAAC,CAAC,IAAI,CAAC,EAAE,kBAAkB,CAAC,CAAC;IAC9F,OAAO,OAAO,CAAC;AACjB,CAAC"}

package/dist/temporal/activities/runtime.d.ts

@@ -0,0 +1,38 @@
+import type { ArtifactStore } from "../../artifacts.js";
+import { type ArtifactRef, type AuthoredTask } from "../../contracts.js";
+import { type ProvenanceMessage } from "../../provenance.js";
+import { type SandboxResult, type SandboxRunOptions } from "../../sandbox/index.js";
+/**
+ * Runs a round sandbox. A provider failure after the wrapper already finished (its status file
+ * was collected) is downgraded to a result carrying the wrapper's status, with the provider's
+ * complaint appended to stderr; any other provider failure stores the partial log and rethrows.
+ */
+export declare function runSandboxWithFailureLog(store: ArtifactStore, logKey: string, action: () => Promise<SandboxResult>): Promise<SandboxResult>;
+export declare function withActivityHeartbeats<T>(detail: string, action: (options: SandboxRunOptions & {
+    readonly signal: AbortSignal;
+}) => Promise<T>): Promise<T>;
+export declare function withTaskBundle<T>(store: ArtifactStore, task: AuthoredTask, action: (taskDirectory: string, root: string) => Promise<T>): Promise<T>;
+export declare function withTemporaryDirectory<T>(prefix: string, action: (root: string) => Promise<T>): Promise<T>;
+/** Heartbeats when running inside a Temporal activity; a no-op in unit tests. */
+export declare function safeHeartbeat(detail: string): void;
+export declare function parseProvenance(value: Uint8Array): ProvenanceMessage[];
+export declare function readAsset(relativePath: string): Promise<Buffer>;
+export interface StoredPiSession {
+    readonly ref: ArtifactRef;
+    readonly finalMessage?: string;
+    /** Tool calls the assistant made, in order. */
+    readonly toolCalls: readonly string[];
+    /** The provider error that ended the session, when the last assistant turn failed. */
+    readonly providerError?: string;
+}
+/**
+ * Persists a collected pi session file as a run artifact. An absent or malformed file yields
+ * undefined so the caller can report the round rather than fail the activity.
+ */
+export declare function storePiSession(store: ArtifactStore, key: string, bytes: Uint8Array | undefined): Promise<StoredPiSession | undefined>;
+/** Reads the held-out and gold patches from an authored submission or a compiled task root. */
+export declare function readTaskPatches(directory: string): Promise<{
+    readonly testPatch: string;
+    readonly goldPatch: string;
+}>;
+//# sourceMappingURL=runtime.d.ts.map

package/dist/temporal/activities/runtime.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"runtime.d.ts","sourceRoot":"","sources":["../../../src/temporal/activities/runtime.ts"],"names":[],"mappings":"AAKA,OAAO,KAAK,EAAE,aAAa,EAAE,MAAM,oBAAoB,CAAC;AACxD,OAAO,EAAE,KAAK,WAAW,EAAE,KAAK,YAAY,EAAwB,MAAM,oBAAoB,CAAC;AAS/F,OAAO,EAAE,KAAK,iBAAiB,EAA2B,MAAM,qBAAqB,CAAC;AACtF,OAAO,EAEL,KAAK,aAAa,EAClB,KAAK,iBAAiB,EACvB,MAAM,wBAAwB,CAAC;AAGhC;;;;GAIG;AACH,wBAAsB,wBAAwB,CAC5C,KAAK,EAAE,aAAa,EACpB,MAAM,EAAE,MAAM,EACd,MAAM,EAAE,MAAM,OAAO,CAAC,aAAa,CAAC,GACnC,OAAO,CAAC,aAAa,CAAC,CAsBxB;AACD,wBAAsB,sBAAsB,CAAC,CAAC,EAC5C,MAAM,EAAE,MAAM,EACd,MAAM,EAAE,CAAC,OAAO,EAAE,iBAAiB,GAAG;IAAE,QAAQ,CAAC,MAAM,EAAE,WAAW,CAAA;CAAE,KAAK,OAAO,CAAC,CAAC,CAAC,GACpF,OAAO,CAAC,CAAC,CAAC,CAsCZ;AACD,wBAAsB,cAAc,CAAC,CAAC,EACpC,KAAK,EAAE,aAAa,EACpB,IAAI,EAAE,YAAY,EAClB,MAAM,EAAE,CAAC,aAAa,EAAE,MAAM,EAAE,IAAI,EAAE,MAAM,KAAK,OAAO,CAAC,CAAC,CAAC,GAC1D,OAAO,CAAC,CAAC,CAAC,CAYZ;AACD,wBAAsB,sBAAsB,CAAC,CAAC,EAC5C,MAAM,EAAE,MAAM,EACd,MAAM,EAAE,CAAC,IAAI,EAAE,MAAM,KAAK,OAAO,CAAC,CAAC,CAAC,GACnC,OAAO,CAAC,CAAC,CAAC,CAOZ;AACD,iFAAiF;AACjF,wBAAgB,aAAa,CAAC,MAAM,EAAE,MAAM,GAAG,IAAI,CAMlD;AACD,wBAAgB,eAAe,CAAC,KAAK,EAAE,UAAU,GAAG,iBAAiB,EAAE,CAMtE;AACD,wBAAgB,SAAS,CAAC,YAAY,EAAE,MAAM,GAAG,OAAO,CAAC,MAAM,CAAC,CAE/D;AACD,MAAM,WAAW,eAAe;IAC9B,QAAQ,CAAC,GAAG,EAAE,WAAW,CAAC;IAC1B,QAAQ,CAAC,YAAY,CAAC,EAAE,MAAM,CAAC;IAC/B,+CAA+C;IAC/C,QAAQ,CAAC,SAAS,EAAE,SAAS,MAAM,EAAE,CAAC;IACtC,sFAAsF;IACtF,QAAQ,CAAC,aAAa,CAAC,EAAE,MAAM,CAAC;CACjC;AACD;;;GAGG;AACH,wBAAsB,cAAc,CAClC,KAAK,EAAE,aAAa,EACpB,GAAG,EAAE,MAAM,EACX,KAAK,EAAE,UAAU,GAAG,SAAS,GAC5B,OAAO,CAAC,eAAe,GAAG,SAAS,CAAC,CAkBtC;AACD,+FAA+F;AAC/F,wBAAsB,eAAe,CACnC,SAAS,EAAE,MAAM,GAChB,OAAO,CAAC;IAAE,QAAQ,CAAC,SAAS,EAAE,MAAM,CAAC;IAAC,QAAQ,CAAC,SAAS,EAAE,MAAM,CAAA;CAAE,CAAC,CAUrE"}

package/dist/temporal/activities/runtime.js

@@ -0,0 +1,146 @@
+import { mkdtemp, readFile, rm, writeFile } from "node:fs/promises";
+import { tmpdir } from "node:os";
+import { join } from "node:path";
+import { CancelledFailure, Context } from "@temporalio/activity";
+import { extractRegularArchive } from "../../archive.js";
+import { taskDefinitionSchema } from "../../contracts.js";
+import { refreshHarborTask } from "../../harbor-task.js";
+import { assertPiSessionFile, finalAssistantMessage, sessionProviderError, toolCallNames, } from "../../pi-session.js";
+import { projectRoot } from "../../project-paths.js";
+import { provenanceMessageSchema } from "../../provenance.js";
+import { SandboxExecutionError, } from "../../sandbox/index.js";
+import { wrapperStatusFrom } from "./round-outcome.js";
+/**
+ * Runs a round sandbox. A provider failure after the wrapper already finished (its status file
+ * was collected) is downgraded to a result carrying the wrapper's status, with the provider's
+ * complaint appended to stderr; any other provider failure stores the partial log and rethrows.
+ */
+export async function runSandboxWithFailureLog(store, logKey, action) {
+    try {
+        return await action();
+    }
+    catch (error) {
+        if (!(error instanceof SandboxExecutionError)) {
+            throw error;
+        }
+        const status = wrapperStatusFrom(error.result.outputs);
+        if (status !== undefined) {
+            return {
+                ...error.result,
+                exitCode: status,
+                stderr: `${error.result.stderr}\n[selfbench] provider failed after the wrapper finished with status ${status}: ${error.message}\n`,
+            };
+        }
+        const log = await store.put(logKey, Buffer.from(`${error.result.stdout}\n${error.result.stderr}`), "text/plain");
+        throw new Error(`${error.message}; partial log: ${log.uri}`, { cause: error });
+    }
+}
+export async function withActivityHeartbeats(detail, action) {
+    const context = Context.current();
+    let outputBytes = 0;
+    let lastOutputAt;
+    const heartbeatDetail = () => ({
+        detail,
+        outputBytes,
+        ...(lastOutputAt ? { lastOutputAt } : {}),
+    });
+    context.heartbeat(heartbeatDetail());
+    const heartbeat = setInterval(() => context.heartbeat(heartbeatDetail()), 60_000);
+    heartbeat.unref();
+    try {
+        try {
+            const result = await action({
+                signal: context.cancellationSignal,
+                onProgress: (progress) => {
+                    outputBytes += progress.bytes;
+                    lastOutputAt = new Date().toISOString();
+                },
+            });
+            if (context.cancellationSignal.aborted) {
+                throw new CancelledFailure("activity cancellation requested");
+            }
+            return result;
+        }
+        catch (error) {
+            if (context.cancellationSignal.aborted && !(error instanceof CancelledFailure)) {
+                throw new CancelledFailure("activity cancellation requested");
+            }
+            throw error;
+        }
+    }
+    finally {
+        clearInterval(heartbeat);
+    }
+}
+export async function withTaskBundle(store, task, action) {
+    return await withTemporaryDirectory(`selfbench-${task.taskId}-`, async (root) => {
+        const archive = join(root, "task.tar.gz");
+        await writeFile(archive, await store.get(task.bundle));
+        await extractRegularArchive(archive, root);
+        const taskDirectory = join(root, "harbor-task");
+        const definition = taskDefinitionSchema.parse(JSON.parse(Buffer.from(await store.get(task.definition)).toString("utf8")));
+        await refreshHarborTask(taskDirectory, definition);
+        return await action(taskDirectory, root);
+    });
+}
+export async function withTemporaryDirectory(prefix, action) {
+    const root = await mkdtemp(join(tmpdir(), prefix));
+    try {
+        return await action(root);
+    }
+    finally {
+        await rm(root, { recursive: true, force: true });
+    }
+}
+/** Heartbeats when running inside a Temporal activity; a no-op in unit tests. */
+export function safeHeartbeat(detail) {
+    try {
+        Context.current().heartbeat(detail);
+    }
+    catch {
+        // Not running inside a Temporal activity; heartbeats are best-effort.
+    }
+}
+export function parseProvenance(value) {
+    return Buffer.from(value)
+        .toString("utf8")
+        .split("\n")
+        .filter((line) => line.trim())
+        .map((line) => provenanceMessageSchema.parse(JSON.parse(line)));
+}
+export function readAsset(relativePath) {
+    return readFile(join(projectRoot(import.meta.url), relativePath));
+}
+/**
+ * Persists a collected pi session file as a run artifact. An absent or malformed file yields
+ * undefined so the caller can report the round rather than fail the activity.
+ */
+export async function storePiSession(store, key, bytes) {
+    if (!bytes || bytes.length === 0) {
+        return undefined;
+    }
+    try {
+        assertPiSessionFile(bytes);
+    }
+    catch {
+        return undefined;
+    }
+    const ref = await store.put(key, bytes, "application/x-ndjson");
+    const finalMessage = finalAssistantMessage(bytes);
+    const providerError = sessionProviderError(bytes);
+    return {
+        ref,
+        toolCalls: toolCallNames(bytes),
+        ...(finalMessage ? { finalMessage } : {}),
+        ...(providerError ? { providerError } : {}),
+    };
+}
+/** Reads the held-out and gold patches from an authored submission or a compiled task root. */
+export async function readTaskPatches(directory) {
+    const [testPatch, goldPatch] = await Promise.all([
+        readFile(join(directory, "test.patch"), "utf8").catch(() => readFile(join(directory, "tests/test.patch"), "utf8")),
+        readFile(join(directory, "gold.patch"), "utf8").catch(() => readFile(join(directory, "solution/gold.patch"), "utf8")),
+    ]);
+    return { testPatch, goldPatch };
+}
+//# sourceMappingURL=runtime.js.map

package/dist/temporal/activities/runtime.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"runtime.js","sourceRoot":"","sources":["../../../src/temporal/activities/runtime.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,OAAO,EAAE,QAAQ,EAAE,EAAE,EAAE,SAAS,EAAE,MAAM,kBAAkB,CAAC;AACpE,OAAO,EAAE,MAAM,EAAE,MAAM,SAAS,CAAC;AACjC,OAAO,EAAE,IAAI,EAAE,MAAM,WAAW,CAAC;AACjC,OAAO,EAAE,gBAAgB,EAAE,OAAO,EAAE,MAAM,sBAAsB,CAAC;AACjE,OAAO,EAAE,qBAAqB,EAAE,MAAM,kBAAkB,CAAC;AAEzD,OAAO,EAAuC,oBAAoB,EAAE,MAAM,oBAAoB,CAAC;AAC/F,OAAO,EAAE,iBAAiB,EAAE,MAAM,sBAAsB,CAAC;AACzD,OAAO,EACL,mBAAmB,EACnB,qBAAqB,EACrB,oBAAoB,EACpB,aAAa,GACd,MAAM,qBAAqB,CAAC;AAC7B,OAAO,EAAE,WAAW,EAAE,MAAM,wBAAwB,CAAC;AACrD,OAAO,EAA0B,uBAAuB,EAAE,MAAM,qBAAqB,CAAC;AACtF,OAAO,EACL,qBAAqB,GAGtB,MAAM,wBAAwB,CAAC;AAChC,OAAO,EAAE,iBAAiB,EAAE,MAAM,oBAAoB,CAAC;AAEvD;;;;GAIG;AACH,MAAM,CAAC,KAAK,UAAU,wBAAwB,CAC5C,KAAoB,EACpB,MAAc,EACd,MAAoC;IAEpC,IAAI,CAAC;QACH,OAAO,MAAM,MAAM,EAAE,CAAC;IACxB,CAAC;IAAC,OAAO,KAAK,EAAE,CAAC;QACf,IAAI,CAAC,CAAC,KAAK,YAAY,qBAAqB,CAAC,EAAE,CAAC;YAC9C,MAAM,KAAK,CAAC;QACd,CAAC;QACD,MAAM,MAAM,GAAG,iBAAiB,CAAC,KAAK,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC;QACvD,IAAI,MAAM,KAAK,SAAS,EAAE,CAAC;YACzB,OAAO;gBACL,GAAG,KAAK,CAAC,MAAM;gBACf,QAAQ,EAAE,MAAM;gBAChB,MAAM,EAAE,GAAG,KAAK,CAAC,MAAM,CAAC,MAAM,wEAAwE,MAAM,KAAK,KAAK,CAAC,OAAO,IAAI;aACnI,CAAC;QACJ,CAAC;QACD,MAAM,GAAG,GAAG,MAAM,KAAK,CAAC,GAAG,CACzB,MAAM,EACN,MAAM,CAAC,IAAI,CAAC,GAAG,KAAK,CAAC,MAAM,CAAC,MAAM,KAAK,KAAK,CAAC,MAAM,CAAC,MAAM,EAAE,CAAC,EAC7D,YAAY,CACb,CAAC;QACF,MAAM,IAAI,KAAK,CAAC,GAAG,KAAK,CAAC,OAAO,kBAAkB,GAAG,CAAC,GAAG,EAAE,EAAE,EAAE,KAAK,EAAE,KAAK,EAAE,CAAC,CAAC;IACjF,CAAC;AACH,CAAC;AACD,MAAM,CAAC,KAAK,UAAU,sBAAsB,CAC1C,MAAc,EACd,MAAqF;IAErF,MAAM,OAAO,GAAG,OAAO,CAAC,OAAO,EAAE,CAAC;IAClC,IAAI,WAAW,GAAG,CAAC,CAAC;IACpB,IAAI,YAAgC,CAAC;IACrC,MAAM,eAAe,GAAG,GAItB,EAAE,CAAC,CAAC;QACJ,MAAM;QACN,WAAW;QACX,GAAG,CAAC,YAAY,CAAC,CAAC,CAAC,EAAE,YAAY,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;KAC1C,CAAC,CAAC;IACH,OAAO,CAAC,SAAS,CAAC,eAAe,EAAE,CAAC,CAAC;IACrC,MAAM,SAAS,GAAG,WAAW,CAAC,GAAG,EAAE,CAAC,OAAO,CAAC,SAAS,CAAC,eAAe,EAAE,CAAC,EAAE,MAAM,CAAC,CAAC;IAClF,SAAS,CAAC,KAAK,EAAE,CAAC;IAClB,IAAI,CAAC;QACH,IAAI,CAAC;YACH,MAAM,MAAM,GAAG,MAAM,MAAM,CAAC;gBAC1B,MAAM,EAAE,OAAO,CAAC,kBAAkB;gBAClC,UAAU,EAAE,CAAC,QAAQ,EAAE,EAAE;oBACvB,WAAW,IAAI,QAAQ,CAAC,KAAK,CAAC;oBAC9B,YAAY,GAAG,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE,CAAC;gBAC1C,CAAC;aACF,CAAC,CAAC;YACH,IAAI,OAAO,CAAC,kBAAkB,CAAC,OAAO,EAAE,CAAC;gBACvC,MAAM,IAAI,gBAAgB,CAAC,iCAAiC,CAAC,CAAC;YAChE,CAAC;YACD,OAAO,MAAM,CAAC;QAChB,CAAC;QAAC,OAAO,KAAK,EAAE,CAAC;YACf,IAAI,OAAO,CAAC,kBAAkB,CAAC,OAAO,IAAI,CAAC,CAAC,KAAK,YAAY,gBAAgB,CAAC,EAAE,CAAC;gBAC/E,MAAM,IAAI,gBAAgB,CAAC,iCAAiC,CAAC,CAAC;YAChE,CAAC;YACD,MAAM,KAAK,CAAC;QACd,CAAC;IACH,CAAC;YAAS,CAAC;QACT,aAAa,CAAC,SAAS,CAAC,CAAC;IAC3B,CAAC;AACH,CAAC;AACD,MAAM,CAAC,KAAK,UAAU,cAAc,CAClC,KAAoB,EACpB,IAAkB,EAClB,MAA2D;IAE3D,OAAO,MAAM,sBAAsB,CAAC,aAAa,IAAI,CAAC,MAAM,GAAG,EAAE,KAAK,EAAE,IAAI,EAAE,EAAE;QAC9E,MAAM,OAAO,GAAG,IAAI,CAAC,IAAI,EAAE,aAAa,CAAC,CAAC;QAC1C,MAAM,SAAS,CAAC,OAAO,EAAE,MAAM,KAAK,CAAC,GAAG,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC,CAAC;QACvD,MAAM,qBAAqB,CAAC,OAAO,EAAE,IAAI,CAAC,CAAC;QAC3C,MAAM,aAAa,GAAG,IAAI,CAAC,IAAI,EAAE,aAAa,CAAC,CAAC;QAChD,MAAM,UAAU,GAAG,oBAAoB,CAAC,KAAK,CAC3C,IAAI,CAAC,KAAK,CAAC,MAAM,CAAC,IAAI,CAAC,MAAM,KAAK,CAAC,GAAG,CAAC,IAAI,CAAC,UAAU,CAAC,CAAC,CAAC,QAAQ,CAAC,MAAM,CAAC,CAAC,CAC3E,CAAC;QACF,MAAM,iBAAiB,CAAC,aAAa,EAAE,UAAU,CAAC,CAAC;QACnD,OAAO,MAAM,MAAM,CAAC,aAAa,EAAE,IAAI,CAAC,CAAC;IAC3C,CAAC,CAAC,CAAC;AACL,CAAC;AACD,MAAM,CAAC,KAAK,UAAU,sBAAsB,CAC1C,MAAc,EACd,MAAoC;IAEpC,MAAM,IAAI,GAAG,MAAM,OAAO,CAAC,IAAI,CAAC,MAAM,EAAE,EAAE,MAAM,CAAC,CAAC,CAAC;IACnD,IAAI,CAAC;QACH,OAAO,MAAM,MAAM,CAAC,IAAI,CAAC,CAAC;IAC5B,CAAC;YAAS,CAAC;QACT,MAAM,EAAE,CAAC,IAAI,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,KAAK,EAAE,IAAI,EAAE,CAAC,CAAC;IACnD,CAAC;AACH,CAAC;AACD,iFAAiF;AACjF,MAAM,UAAU,aAAa,CAAC,MAAc;IAC1C,IAAI,CAAC;QACH,OAAO,CAAC,OAAO,EAAE,CAAC,SAAS,CAAC,MAAM,CAAC,CAAC;IACtC,CAAC;IAAC,MAAM,CAAC;QACP,sEAAsE;IACxE,CAAC;AACH,CAAC;AACD,MAAM,UAAU,eAAe,CAAC,KAAiB;IAC/C,OAAO,MAAM,CAAC,IAAI,CAAC,KAAK,CAAC;SACtB,QAAQ,CAAC,MAAM,CAAC;SAChB,KAAK,CAAC,IAAI,CAAC;SACX,MAAM,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,IAAI,EAAE,CAAC;SAC7B,GAAG,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,uBAAuB,CAAC,KAAK,CAAC,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;AACpE,CAAC;AACD,MAAM,UAAU,SAAS,CAAC,YAAoB;IAC5C,OAAO,QAAQ,CAAC,IAAI,CAAC,WAAW,CAAC,MAAM,CAAC,IAAI,CAAC,GAAG,CAAC,EAAE,YAAY,CAAC,CAAC,CAAC;AACpE,CAAC;AASD;;;GAGG;AACH,MAAM,CAAC,KAAK,UAAU,cAAc,CAClC,KAAoB,EACpB,GAAW,EACX,KAA6B;IAE7B,IAAI,CAAC,KAAK,IAAI,KAAK,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QACjC,OAAO,SAAS,CAAC;IACnB,CAAC;IACD,IAAI,CAAC;QACH,mBAAmB,CAAC,KAAK,CAAC,CAAC;IAC7B,CAAC;IAAC,MAAM,CAAC;QACP,OAAO,SAAS,CAAC;IACnB,CAAC;IACD,MAAM,GAAG,GAAG,MAAM,KAAK,CAAC,GAAG,CAAC,GAAG,EAAE,KAAK,EAAE,sBAAsB,CAAC,CAAC;IAChE,MAAM,YAAY,GAAG,qBAAqB,CAAC,KAAK,CAAC,CAAC;IAClD,MAAM,aAAa,GAAG,oBAAoB,CAAC,KAAK,CAAC,CAAC;IAClD,OAAO;QACL,GAAG;QACH,SAAS,EAAE,aAAa,CAAC,KAAK,CAAC;QAC/B,GAAG,CAAC,YAAY,CAAC,CAAC,CAAC,EAAE,YAAY,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;QACzC,GAAG,CAAC,aAAa,CAAC,CAAC,CAAC,EAAE,aAAa,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;KAC5C,CAAC;AACJ,CAAC;AACD,+FAA+F;AAC/F,MAAM,CAAC,KAAK,UAAU,eAAe,CACnC,SAAiB;IAEjB,MAAM,CAAC,SAAS,EAAE,SAAS,CAAC,GAAG,MAAM,OAAO,CAAC,GAAG,CAAC;QAC/C,QAAQ,CAAC,IAAI,CAAC,SAAS,EAAE,YAAY,CAAC,EAAE,MAAM,CAAC,CAAC,KAAK,CAAC,GAAG,EAAE,CACzD,QAAQ,CAAC,IAAI,CAAC,SAAS,EAAE,kBAAkB,CAAC,EAAE,MAAM,CAAC,CACtD;QACD,QAAQ,CAAC,IAAI,CAAC,SAAS,EAAE,YAAY,CAAC,EAAE,MAAM,CAAC,CAAC,KAAK,CAAC,GAAG,EAAE,CACzD,QAAQ,CAAC,IAAI,CAAC,SAAS,EAAE,qBAAqB,CAAC,EAAE,MAAM,CAAC,CACzD;KACF,CAAC,CAAC;IACH,OAAO,EAAE,SAAS,EAAE,SAAS,EAAE,CAAC;AAClC,CAAC"}

package/dist/temporal/activities/session-verify.d.ts

@@ -0,0 +1,45 @@
+import type { ArtifactStore } from "../../artifacts.js";
+import type { SelfBenchConfig } from "../../config.js";
+import { type ArtifactRef, type AuthoredTask, type Candidate, type RunRequest, type VerifyStage } from "../../contracts.js";
+import type { LiveSandbox } from "../../sandbox/index.js";
+import { type MailboxRequest, type MailboxResponse } from "../../sandbox/supervisor.js";
+import { type OriginalTask } from "./drafts.js";
+export interface SessionVerifyContext {
+    readonly store: ArtifactStore;
+    readonly harborEnvironment: SelfBenchConfig["harborEnvironment"];
+    readonly run: RunRequest;
+    readonly candidate: Candidate;
+    readonly stage: VerifyStage;
+    readonly round: number;
+    /** Round artifact prefix; verifies are stored under `<prefix>/verify-<k>/`. */
+    readonly prefix: string;
+    /** For verifier sessions: the task a fix must stay within. */
+    readonly original?: OriginalTask;
+}
+export interface SessionVerifyRecord {
+    readonly index: number;
+    readonly hash: string;
+    readonly green: boolean;
+    readonly report: ArtifactRef;
+    readonly task?: AuthoredTask;
+}
+export interface VerifiedSubmission {
+    readonly report: ArtifactRef;
+    readonly task: AuthoredTask;
+}
+/**
+ * Worker side of the agent's `verify` tool: each mailbox request becomes a draft, goes through the
+ * same compile/audit/build/smoke/nop/oracle as a submission, and is archived per call. A later
+ * submission with the same payload hash reuses the green result.
+ */
+export declare class SessionVerifier {
+    #private;
+    readonly records: SessionVerifyRecord[];
+    constructor(context: SessionVerifyContext);
+    /** Supervises the live sandbox mailbox for the duration of the agent command. */
+    supervise(sandbox: LiveSandbox, exited: AbortSignal): Promise<void>;
+    handle(request: MailboxRequest): Promise<MailboxResponse>;
+    /** The green verify matching a submission, if the agent submitted exactly what it verified. */
+    verified(definition: unknown, testPatch: string, goldPatch: string): VerifiedSubmission | undefined;
+}
+//# sourceMappingURL=session-verify.d.ts.map

package/dist/temporal/activities/session-verify.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"session-verify.d.ts","sourceRoot":"","sources":["../../../src/temporal/activities/session-verify.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,EAAE,aAAa,EAAE,MAAM,oBAAoB,CAAC;AACxD,OAAO,KAAK,EAAE,eAAe,EAAE,MAAM,iBAAiB,CAAC;AACvD,OAAO,EACL,KAAK,WAAW,EAChB,KAAK,YAAY,EACjB,KAAK,SAAS,EACd,KAAK,UAAU,EAEf,KAAK,WAAW,EACjB,MAAM,oBAAoB,CAAC;AAC5B,OAAO,KAAK,EAAE,WAAW,EAAE,MAAM,wBAAwB,CAAC;AAC1D,OAAO,EACL,KAAK,cAAc,EACnB,KAAK,eAAe,EAErB,MAAM,6BAA6B,CAAC;AAIrC,OAAO,EAAoB,KAAK,YAAY,EAAE,MAAM,aAAa,CAAC;AAGlE,MAAM,WAAW,oBAAoB;IACnC,QAAQ,CAAC,KAAK,EAAE,aAAa,CAAC;IAC9B,QAAQ,CAAC,iBAAiB,EAAE,eAAe,CAAC,mBAAmB,CAAC,CAAC;IACjE,QAAQ,CAAC,GAAG,EAAE,UAAU,CAAC;IACzB,QAAQ,CAAC,SAAS,EAAE,SAAS,CAAC;IAC9B,QAAQ,CAAC,KAAK,EAAE,WAAW,CAAC;IAC5B,QAAQ,CAAC,KAAK,EAAE,MAAM,CAAC;IACvB,+EAA+E;IAC/E,QAAQ,CAAC,MAAM,EAAE,MAAM,CAAC;IACxB,8DAA8D;IAC9D,QAAQ,CAAC,QAAQ,CAAC,EAAE,YAAY,CAAC;CAClC;AAED,MAAM,WAAW,mBAAmB;IAClC,QAAQ,CAAC,KAAK,EAAE,MAAM,CAAC;IACvB,QAAQ,CAAC,IAAI,EAAE,MAAM,CAAC;IACtB,QAAQ,CAAC,KAAK,EAAE,OAAO,CAAC;IACxB,QAAQ,CAAC,MAAM,EAAE,WAAW,CAAC;IAC7B,QAAQ,CAAC,IAAI,CAAC,EAAE,YAAY,CAAC;CAC9B;AAED,MAAM,WAAW,kBAAkB;IACjC,QAAQ,CAAC,MAAM,EAAE,WAAW,CAAC;IAC7B,QAAQ,CAAC,IAAI,EAAE,YAAY,CAAC;CAC7B;AAED;;;;GAIG;AACH,qBAAa,eAAe;;IAC1B,QAAQ,CAAC,OAAO,EAAE,mBAAmB,EAAE,CAAM;gBAGjC,OAAO,EAAE,oBAAoB;IAIzC,iFAAiF;IACjF,SAAS,CAAC,OAAO,EAAE,WAAW,EAAE,MAAM,EAAE,WAAW,GAAG,OAAO,CAAC,IAAI,CAAC;IAQ7D,MAAM,CAAC,OAAO,EAAE,cAAc,GAAG,OAAO,CAAC,eAAe,CAAC;IAsD/D,+FAA+F;IAC/F,QAAQ,CACN,UAAU,EAAE,OAAO,EACnB,SAAS,EAAE,MAAM,EACjB,SAAS,EAAE,MAAM,GAChB,kBAAkB,GAAG,SAAS;CAOlC"}