fix(opencode): close rejected completion tool activity

Co-Authored-By: Paperclip <noreply@paperclip.ing>

packages/paperclip-runner/src/drivers/opencode/opencode-server-driver.test.ts

@@ -21,6 +21,7 @@ import {
   nativeRuntimePromptDigest,
   type NativeRuntimeContextSnapshot,
 } from "../../contracts/runtime-context.js";
+import { createCodexTaskEnvelope } from "../../contracts/codex.js";
 import { localIntegrityBoundaryGolden } from "../../../test-support/local-integrity-boundary-golden.js";
 import {
   OpenCodeServerDriver,
@@ -1455,23 +1456,25 @@ describe("OpenCodeServerDriver", () => {
     ).resolves.toMatchObject({ ok: true });
   });
 
-  it("returns a repairable tool error for wrong criteria without committing a bad semantic result", async () => {
+  it.each(["wrong", "missing", "duplicate"])("returns a repairable tool error for %s criteria without committing a bad semantic result", async (mode) => {
     await chmod(fixture, 0o755);
     const root = await mkdtemp(join(tmpdir(), "paperclip-opencode-criteria-"));
     const workspace = await mkdtemp(join(tmpdir(), "paperclip-opencode-workspace-"));
     roots.push(root, workspace);
-    const driver = new OpenCodeServerDriver({ model: "openrouter/deepseek/deepseek-v4-flash-0731", runtimeDirectory: root, command: fixture, environment: { PATH: process.env.PATH, OPENROUTER_API_KEY: "fixture-key" } });
+    const driver = new OpenCodeServerDriver({ model: "openrouter/deepseek/deepseek-v4-flash-0731", runtimeDirectory: root, command: fixture, environment: { PATH: process.env.PATH, OPENROUTER_API_KEY: "fixture-key" }, taskEnvelope: createCodexTaskEnvelope({ objective: "Apply the accepted decision", contractRevision: "approval-v2", criteria: [{ id: "human_response", requirement: "Apply the approved response" }] }) });
     const session = await driver.openSession({ runId: "criteria-repair", normalizedSessionId: "criteria-repair", workingDirectory: workspace });
-    await session.startTurn({ message: { role: "user", text: "repair-criteria" } });
+    await session.startTurn({ message: { role: "user", text: `repair-criteria-${mode}` } });
     const events = [];
     for await (const event of session.events()) events.push(event);
     const results = events.filter((event) => event.eventType === "run.result.proposed");
     expect(results).toHaveLength(1);
-    expect(results[0].payload).toMatchObject({ completionClaim: { criteria: [{ criterionId: "objective" }] } });
+    expect(events.some((event) => event.eventType === "item.completed" &&
+      (event.payload as { item?: { is_error?: boolean } }).item?.is_error === true)).toBe(true);
+    expect(results[0].payload).toMatchObject({ completionClaim: { criteria: [{ criterionId: "human_response" }] } });
     const files = await readdir(root, { recursive: true });
     const evidence = files.find((name) => name.endsWith("fake-criteria-repair.json"));
     expect(evidence).toBeDefined();
-    expect(JSON.parse(await readFile(join(root, evidence!), "utf8"))).toMatchObject({ result: { isError: true, content: [{ text: expect.stringContaining('"objective"') }] } });
+    expect(JSON.parse(await readFile(join(root, evidence!), "utf8"))).toMatchObject({ result: { isError: true, content: [{ text: expect.stringContaining('"human_response"') }] } });
     await session.close({ reason: "test" });
   });
 

packages/paperclip-runner/src/drivers/opencode/opencode-server-driver.ts

@@ -887,60 +887,71 @@ class OpenCodeHarnessSession implements HarnessSession {
       { turnId, itemId: call.callId },
     );
     if (tool === PRP_COMPLETION_TOOL_NAME || tool === PRP_BLOCK_TOOL_NAME) {
-      const validation = validatePrpStructuredRunResult(call.arguments);
-      if (!validation.ok) throw new Error("Invalid semantic result");
-      if (
-        (tool === PRP_BLOCK_TOOL_NAME &&
-          validation.result.reportedWorkDisposition !== "blocked") ||
-        (tool === PRP_COMPLETION_TOOL_NAME &&
-          validation.result.reportedWorkDisposition === "blocked")
-      )
-        throw new Error(
-          "Semantic result disposition does not match the terminal tool",
-        );
-      if (
-        validation.result.completionClaim.contractRevision !==
-        this.#taskEnvelope.completionContract.revision
-      ) {
-        throw new Error(
-          "Semantic result completion contract revision does not match",
-        );
-      }
-      const expectedIds = this.#taskEnvelope.completionContract.criteria.map((criterion) => criterion.id);
-      const receivedIds = validation.result.completionClaim.criteria.map((criterion) => criterion.criterionId);
-      if (receivedIds.length !== expectedIds.length || new Set(receivedIds).size !== receivedIds.length || receivedIds.some((id) => !expectedIds.includes(id))) {
-        // Reject at the tool boundary so the provider can repair its claim.
-        // Emitting a bad result here kills runnerd's strict outer validation.
-        throw new Error(`Semantic result criteria must contain exactly these criterionIds, once each: ${JSON.stringify(expectedIds)}. Keep contractRevision ${JSON.stringify(this.#taskEnvelope.completionContract.revision)}.`);
-      }
-      const fingerprint = canonicalJson(validation.result);
-      if (this.#resultFingerprint && this.#resultFingerprint !== fingerprint)
-        throw new Error("A different semantic result was already committed");
-      if (!this.#resultFingerprint) {
-        this.#result = structuredClone(validation.result);
-        this.#resultFingerprint = fingerprint;
-        this.#resultCallId = call.callId;
-        this.#resultTurnId = turnId;
-        this.#semanticResultTextBoundary = this.#completedTextParts.length;
-        this.#emit("run.result.proposed", validation.result, {
-          turnId,
-          itemId: call.callId,
-        });
-      }
-      this.#emit(
-        "item.completed",
-        {
-          kind: "dynamicToolCall",
-          item: {
-            type: "tool_result",
-            id: call.callId,
-            tool_use_id: call.callId,
-            result: "Semantic completion accepted.",
+      try {
+        const validation = validatePrpStructuredRunResult(call.arguments);
+        if (!validation.ok) throw new Error("Invalid semantic result");
+        if (
+          (tool === PRP_BLOCK_TOOL_NAME &&
+            validation.result.reportedWorkDisposition !== "blocked") ||
+          (tool === PRP_COMPLETION_TOOL_NAME &&
+            validation.result.reportedWorkDisposition === "blocked")
+        )
+          throw new Error(
+            "Semantic result disposition does not match the terminal tool",
+          );
+        if (
+          validation.result.completionClaim.contractRevision !==
+          this.#taskEnvelope.completionContract.revision
+        ) {
+          throw new Error(
+            "Semantic result completion contract revision does not match",
+          );
+        }
+        const expectedIds = this.#taskEnvelope.completionContract.criteria.map((criterion) => criterion.id);
+        const receivedIds = validation.result.completionClaim.criteria.map((criterion) => criterion.criterionId);
+        if (receivedIds.length !== expectedIds.length || new Set(receivedIds).size !== receivedIds.length || receivedIds.some((id) => !expectedIds.includes(id))) {
+          // Reject at the tool boundary so the provider can repair its claim.
+          // Emitting a bad result here kills runnerd's strict outer validation.
+          throw new Error(`Semantic result criteria must contain exactly these criterionIds, once each: ${JSON.stringify(expectedIds)}. Keep contractRevision ${JSON.stringify(this.#taskEnvelope.completionContract.revision)}.`);
+        }
+        const fingerprint = canonicalJson(validation.result);
+        if (this.#resultFingerprint && this.#resultFingerprint !== fingerprint)
+          throw new Error("A different semantic result was already committed");
+        if (!this.#resultFingerprint) {
+          this.#result = structuredClone(validation.result);
+          this.#resultFingerprint = fingerprint;
+          this.#resultCallId = call.callId;
+          this.#resultTurnId = turnId;
+          this.#semanticResultTextBoundary = this.#completedTextParts.length;
+          this.#emit("run.result.proposed", validation.result, {
+            turnId,
+            itemId: call.callId,
+          });
+        }
+        this.#emit(
+          "item.completed",
+          {
+            kind: "dynamicToolCall",
+            item: {
+              type: "tool_result",
+              id: call.callId,
+              tool_use_id: call.callId,
+              result: "Semantic completion accepted.",
+            },
           },
-        },
-        { turnId, itemId: call.callId },
-      );
-      return { accepted: true };
+          { turnId, itemId: call.callId },
+        );
+        return { accepted: true };
+      } catch (error) {
+        // A rejected semantic call still completes its tool activity item.
+        // Otherwise a later question can appear to have an in-flight tool.
+        this.#emit("item.completed", {
+          kind: "dynamicToolCall",
+          item: { type: "tool_result", id: call.callId, tool_use_id: call.callId,
+            is_error: true, error: error instanceof Error ? error.message : String(error) },
+        }, { turnId, itemId: call.callId });
+        throw error;
+      }
     }
     if (!this.#dynamicToolHandler)
       throw new Error("Unsupported Paperclip operation");

packages/paperclip-runner/test/fixtures/fake-opencode-server.mjs

@@ -206,7 +206,9 @@ async function callTerminalTool(promptBody) {
   };
   if (String(prompt.message ?? "").includes("repair-criteria")) {
     const bad = structuredClone(result);
-    bad.completionClaim.criteria = [{ criterionId: "invented", status: "satisfied", evidenceRefs: [] }];
+    bad.completionClaim.criteria = String(prompt.message).includes("repair-criteria-missing") ? []
+      : String(prompt.message).includes("repair-criteria-duplicate") ? [...result.completionClaim.criteria, ...result.completionClaim.criteria]
+      : [{ criterionId: "invented", status: "satisfied", evidenceRefs: [] }];
     const rejected = await mcpRequest("tools/call", { name: "paperclip_finish", arguments: bad });
     await writeFile(join(process.env.XDG_DATA_HOME, "fake-criteria-repair.json"), JSON.stringify(rejected));
   }