fix: finalize v4 remediation closure

Her-xanadu · sisyphus-dev-ai · Her-xanadu · commit c0089180639f · 2026-03-16T19:32:22.000+08:00
Ultraworked with [Sisyphus](https://github.com/code-yeongyu/oh-my-opencode) Co-authored-by: Sisyphus <clio-agent@sisyphuslabs.ai>
diff --git a/fixtures/results/result_packet-good.json b/fixtures/results/result_packet-good.json
diff --git a/fixtures/runs/crash-after-checkpoint/experiments/recovery_checkpoint.json b/fixtures/runs/crash-after-checkpoint/experiments/recovery_checkpoint.json
@@ -0,0 +1,5 @@
+{
+  "run_id": "run-crash",
+  "checkpoint_path": "checkpoints/latest.ckpt",
+  "parent_run_id": "run-parent"
+}
diff --git a/scripts/docker/scientific-task-smoke.ts b/scripts/docker/scientific-task-smoke.ts
@@ -13,13 +13,13 @@ import {
 } from "../../src/tools";
 import { pathExists, readJsonl } from "../../src/utils/fs";
 import {
-  getCompatAttemptsPath,
-  getCompatBestPath,
   getCompatGoalPath,
-  getCompatProposalCardsPath,
-  getCompatSessionPath,
+  getBestPath,
   getOrchestrationSummaryPath,
   getOrchestrationTracePath,
+  getProposalCardsPath,
+  getRunsPath,
+  getSessionPath,
 } from "../../src/utils/paths";
 
 async function writeToyResearchWorkspace(workspaceRoot: string): Promise<void> {
@@ -97,13 +97,13 @@ async function main() {
   const acceptance = JSON.parse(await experiment_acceptance_review.execute({ workspace_root: workspaceRoot }));
 
   const trace = await readJsonl<{ actor: string }>(getOrchestrationTracePath(workspaceRoot));
-  const attempts = await readJsonl<{ run_id: string; current_metric: number; status: string }>(getCompatAttemptsPath(workspaceRoot));
-  const proposalCards = await readJsonl(getCompatProposalCardsPath(workspaceRoot));
+  const attempts = await readJsonl<{ run_id: string; current_metric: number; status: string }>(getRunsPath(workspaceRoot));
+  const proposalCards = await readJsonl(getProposalCardsPath(workspaceRoot));
 
   const artifactPresence = {
     compat_goal: await pathExists(getCompatGoalPath(workspaceRoot)),
-    compat_session: await pathExists(getCompatSessionPath(workspaceRoot)),
-    compat_best: await pathExists(getCompatBestPath(workspaceRoot)),
+    session: await pathExists(getSessionPath(workspaceRoot)),
+    best: await pathExists(getBestPath(workspaceRoot)),
     orchestration_summary: await pathExists(getOrchestrationSummaryPath(workspaceRoot)),
   };
 
@@ -125,7 +125,7 @@ async function main() {
       throw new Error(`missing orchestration actor: ${expectedActor}`);
     }
   }
-  if (!artifactPresence.compat_goal || !artifactPresence.compat_session || !artifactPresence.compat_best || !artifactPresence.orchestration_summary) {
+  if (!artifactPresence.compat_goal || !artifactPresence.session || !artifactPresence.best || !artifactPresence.orchestration_summary) {
     throw new Error(`artifact presence check failed: ${JSON.stringify(artifactPresence)}`);
   }
 
diff --git a/scripts/kb/run_inference_cycle.py b/scripts/kb/run_inference_cycle.py
@@ -39,11 +39,6 @@ def main() -> None:
     best = workspace_root / "experiments" / "best.json"
     attempts = workspace_root / "experiments" / "attempts.jsonl"
     result_packet = workspace_root / "experiments" / "result_packet.json"
-    if not result_packet.exists():
-        fallback_result_packet = workspace_root / "experiments" / "result-packet.json"
-        result_packet = (
-            fallback_result_packet if fallback_result_packet.exists() else result_packet
-        )
     if not (session.exists() and best.exists() and attempts.exists()):
         emit_json(
             {
diff --git a/scripts/run_candidate.py b/scripts/run_candidate.py
@@ -21,6 +21,7 @@
     run_dir,
     run_stage,
     restore_parent_snapshot,
+    read_json,
     save_parent_snapshot,
     save_run_manifest,
     save_pending_result,
@@ -144,6 +145,23 @@ def main() -> None:
             "parent_run_id": current_best_exp_ref(workspace) or "baseline",
         },
     )
+    parent_snapshot = read_json(
+        workspace / "experiments" / "runs" / run_id / "parent_snapshot.json", {}
+    )
+    created_files = [
+        relative
+        for relative, entry in parent_snapshot.items()
+        if isinstance(entry, dict)
+        and not entry.get("exists", False)
+        and (workspace / relative).exists()
+    ]
+    deleted_files = [
+        relative
+        for relative, entry in parent_snapshot.items()
+        if isinstance(entry, dict)
+        and entry.get("exists", False)
+        and not (workspace / relative).exists()
+    ]
     save_run_manifest(
         workspace,
         run_id,
@@ -152,8 +170,8 @@ def main() -> None:
             "proposal_id": mutation.get("proposal_id"),
             "family": mutation.get("family"),
             "touched_files": touched_files,
-            "created_files": [],
-            "deleted_files": [],
+            "created_files": created_files,
+            "deleted_files": deleted_files,
             "checkpoint_path": str(checkpoint),
             "dvc_exp_ref": run_id,
             "resume_from": args.resume_from,
diff --git a/src/agents/prometheus-bootstrap-planner.ts b/src/agents/prometheus-bootstrap-planner.ts
@@ -1,6 +1,5 @@
 export const prometheusBootstrapPlannerAgent = {
   mode: "primary",
-  model: "openai/gpt-5.4",
   color: "#FF5722",
   description: "Prometheus override used only for bootstrap and review-blocked replanning.",
   permission: {
diff --git a/src/agents/sisyphus-experiment-orchestrator.ts b/src/agents/sisyphus-experiment-orchestrator.ts
@@ -1,6 +1,5 @@
 export const sisyphusExperimentOrchestratorAgent = {
   mode: "primary",
-  model: "openai/gpt-5.4",
   color: "#00CED1",
   description: "Sisyphus override for the fixed experiment loop orchestrator.",
   permission: {
diff --git a/src/experiment/session.ts b/src/experiment/session.ts
@@ -1,5 +1,4 @@
 import type { infer as ZodInfer } from "zod";
-import { syncSessionArtifact } from "../compat/artifacts";
 import { controllerSessionSchema } from "../controller/schema";
 import { readJson, writeJson } from "../utils/fs";
 import { createId } from "../utils/ids";
@@ -41,5 +40,4 @@ export async function loadSession(workspaceRoot: string): Promise<ExperimentSess
 export async function saveSession(workspaceRoot: string, session: ExperimentSession): Promise<void> {
   const parsed = experimentSessionSchema.parse(session);
   await writeJson(getSessionPath(workspaceRoot), parsed);
-  await syncSessionArtifact(workspaceRoot, parsed);
 }
diff --git a/src/loop/execute-iteration.ts b/src/loop/execute-iteration.ts
@@ -1,4 +1,3 @@
-import { syncAttemptArtifact, syncBestArtifact, syncRunEventArtifact } from "../compat/artifacts";
 import { appendJsonl, writeJson } from "../utils/fs";
 import { getBestPath, getResultPacketPath, getRunEventsPath, getRunsPath } from "../utils/paths";
 import { nowIso } from "../utils/time";
@@ -36,7 +35,6 @@ async function writeRunEvent(
     payload,
   };
   await appendJsonl(getRunEventsPath(workspaceRoot, runId), event);
-  await syncRunEventArtifact(workspaceRoot, runId, event);
 }
 
 export async function executeIteration(input: {
@@ -97,7 +95,6 @@ export async function executeIteration(input: {
     status: decision.status,
   };
   await appendJsonl(getRunsPath(input.workspaceRoot), record);
-  await syncAttemptArtifact(input.workspaceRoot, record);
   if (decision.status === "keep") {
     const best = {
       current_best: { run_id: runId, metric: currentMetric, commit: "local", checkpoint: null },
@@ -106,7 +103,6 @@ export async function executeIteration(input: {
       updated_at: nowIso(),
     };
     await writeJson(getBestPath(input.workspaceRoot), best);
-    await syncBestArtifact(input.workspaceRoot, best);
   }
   return {
     run_id: runId,
diff --git a/src/recovery/resume.ts b/src/recovery/resume.ts
@@ -1,4 +1,3 @@
-import path from "node:path";
 import { readJson, writeJson } from "../utils/fs";
 import { getRecoveryCheckpointPath } from "../utils/paths";
 
@@ -13,10 +12,7 @@ export async function saveRecoveryCheckpoint(workspaceRoot: string, checkpoint:
 }
 
 export async function resumeExperiment(workspaceRoot: string): Promise<{ resumed: boolean; source: string | null }> {
-  let checkpoint = await readJson<RecoveryCheckpoint | null>(getRecoveryCheckpointPath(workspaceRoot), null);
-  if (!checkpoint) {
-    checkpoint = await readJson<RecoveryCheckpoint | null>(path.join(workspaceRoot, ".opencode", "auto-experiment", "recovery-checkpoint.json"), null);
-  }
+  const checkpoint = await readJson<RecoveryCheckpoint | null>(getRecoveryCheckpointPath(workspaceRoot), null);
   if (!checkpoint) {
     return { resumed: false, source: null };
   }
diff --git a/src/state/best-store.ts b/src/state/best-store.ts
@@ -1,5 +1,4 @@
 import { z } from "zod";
-import { syncBestArtifact } from "../compat/artifacts";
 import { readJson, writeJson } from "../utils/fs";
 import { getBestPath } from "../utils/paths";
 import { nowIso } from "../utils/time";
@@ -52,7 +51,6 @@ export class BestStoreManager {
   async save(value: BestStore): Promise<BestStore> {
     const parsed = bestStoreSchema.parse({ ...value, updated_at: nowIso() });
     await writeJson(getBestPath(this.workspaceRoot), parsed);
-    await syncBestArtifact(this.workspaceRoot, parsed);
     return parsed;
   }
 }
diff --git a/src/tools/index.ts b/src/tools/index.ts
@@ -2,7 +2,7 @@ import { tool, type PluginInput, type ToolDefinition } from "../opencode-plugin"
 import { aggregateProposals } from "../analysis/aggregator";
 import { resultPacketSchema } from "../analysis/result-packet";
 import { runTriModelAnalysis } from "../analysis/tri-model";
-import { syncGoalArtifact, syncProposalCardArtifact } from "../compat/artifacts";
+import { syncGoalArtifact } from "../compat/artifacts";
 import { buildSession, loadSession, saveSession } from "../experiment/session";
 import { prepareDockerSandbox } from "../sandbox/docker-runner";
 import { sandboxPreparationSchema } from "../sandbox/schema";
@@ -210,7 +210,6 @@ export const experiment_run_analysis: ToolDefinition = tool({
     const cards = runTriModelAnalysis(packet);
     for (const card of cards) {
       await appendJsonl(getProposalCardsPath(workspaceRoot), card);
-      await syncProposalCardArtifact(workspaceRoot, card);
     }
     return stringify(aggregateProposals(cards));
   },
diff --git a/src/utils/paths.ts b/src/utils/paths.ts
@@ -100,11 +100,11 @@ export function getRunEventsPath(workspaceRoot: string, runId: string): string {
 }
 
 export function getRecoveryJournalPath(workspaceRoot: string): string {
-  return path.join(getExperimentRoot(workspaceRoot), "recovery-journal.jsonl");
+  return path.join(getExperimentRoot(workspaceRoot), "recovery_journal.jsonl");
 }
 
 export function getRecoveryCheckpointPath(workspaceRoot: string): string {
-  return path.join(getExperimentRoot(workspaceRoot), "recovery-checkpoint.json");
+  return path.join(getExperimentRoot(workspaceRoot), "recovery_checkpoint.json");
 }
 
 export function resolveWorkspaceRoot(workspaceRoot?: string): string {
diff --git a/tests/analysis/schema.test.ts b/tests/analysis/schema.test.ts
@@ -1,5 +1,5 @@
 import { describe, expect, it } from "vitest";
-import resultPacket from "../../fixtures/results/result-packet-good.json";
+import resultPacket from "../../fixtures/results/result_packet-good.json";
 import { proposalCardSchema, proposalContractSchema } from "../../src/analysis/proposal-card";
 import { controllerSessionSchema, controllerResearchContextSchema, resultPacketSchema } from "../../src/analysis/result-packet";
 import { controllerEvidenceMetadataSchema, controllerRetrievalResultSchema } from "../../src/controller/schema";
diff --git a/tests/analysis/tri-model.test.ts b/tests/analysis/tri-model.test.ts
@@ -1,5 +1,5 @@
 import { describe, expect, it } from "vitest";
-import resultPacket from "../../fixtures/results/result-packet-good.json";
+import resultPacket from "../../fixtures/results/result_packet-good.json";
 import { runTriModelAnalysis } from "../../src/analysis/tri-model";
 import type { ResultPacket } from "../../src/analysis/result-packet";
 
diff --git a/tests/e2e/e2e.test.ts b/tests/e2e/e2e.test.ts
@@ -72,5 +72,5 @@ describe("local e2e", () => {
       ]),
     );
     expect(status.best.current_best.metric).toBeGreaterThan(0.5);
-  }, 45000);
+  }, 60000);
 });
diff --git a/tests/e2e/model-source-consistency.test.ts b/tests/e2e/model-source-consistency.test.ts
@@ -0,0 +1,18 @@
+import fs from "node:fs/promises";
+import path from "node:path";
+import { fileURLToPath } from "node:url";
+import { describe, expect, it } from "vitest";
+
+const repoRoot = path.resolve(path.dirname(fileURLToPath(import.meta.url)), "../..");
+
+describe("model source consistency", () => {
+  it("keeps primary agent defaults in opencode.json instead of hardcoded TS models", async () => {
+    const orchestrator = await fs.readFile(path.join(repoRoot, "src", "agents", "sisyphus-experiment-orchestrator.ts"), "utf8");
+    const planner = await fs.readFile(path.join(repoRoot, "src", "agents", "prometheus-bootstrap-planner.ts"), "utf8");
+    expect(orchestrator).not.toContain('model: "openai/gpt-5.4"');
+    expect(planner).not.toContain('model: "openai/gpt-5.4"');
+
+    const opencodeConfig = JSON.parse(await fs.readFile(path.join(repoRoot, "opencode.json"), "utf8"));
+    expect(opencodeConfig.agent.Apollo.model).toBeTruthy();
+  });
+});
diff --git a/tests/e2e/python-controller.test.ts b/tests/e2e/python-controller.test.ts
@@ -179,7 +179,7 @@ describe("python controller cli", () => {
     const resume = await runInnovationLoop(workspace, configPath, "resume");
     expect(resume.resumed).toBe(true);
     expect(resume.candidate.queued).toBe(true);
-  }, 45000);
+  }, 60000);
 
   it("supports detached start and stop", async () => {
     const { workspace, configPath } = await makeWorkspace();

Original file line number	Diff line number	Diff line change
`@@ -1,4 +1,3 @@`
`1`		`-import path from "node:path";`
`2`	`1`	`import { readJson, writeJson } from "../utils/fs";`
`3`	`2`	`import { getRecoveryCheckpointPath } from "../utils/paths";`
`4`	`3`
`@@ -13,10 +12,7 @@ export async function saveRecoveryCheckpoint(workspaceRoot: string, checkpoint:`
`13`	`12`	`}`
`14`	`13`
`15`	`14`	`export async function resumeExperiment(workspaceRoot: string): Promise<{ resumed: boolean; source: string \| null }> {`
`16`		`- let checkpoint = await readJson<RecoveryCheckpoint \| null>(getRecoveryCheckpointPath(workspaceRoot), null);`
`17`		`- if (!checkpoint) {`
`18`		`- checkpoint = await readJson<RecoveryCheckpoint \| null>(path.join(workspaceRoot, ".opencode", "auto-experiment", "recovery-checkpoint.json"), null);`
`19`		`- }`
	`15`	`+ const checkpoint = await readJson<RecoveryCheckpoint \| null>(getRecoveryCheckpointPath(workspaceRoot), null);`
`20`	`16`	`if (!checkpoint) {`
`21`	`17`	`return { resumed: false, source: null };`
`22`	`18`	`}`