diff --git a/packages/deepsec/src/__tests__/setup-coordinator.test.ts b/packages/deepsec/src/__tests__/setup-coordinator.test.ts
index f3b312c..c772e05 100644
--- a/packages/deepsec/src/__tests__/setup-coordinator.test.ts
+++ b/packages/deepsec/src/__tests__/setup-coordinator.test.ts
@@ -13,6 +13,19 @@ import {
   setupErrorExitCode,
 } from "../setup/protocol.js";
 
+// Stub the osv enrichment so the coordinator path can be exercised without the
+// external osv-scanner/uv/syft toolchain. `runOsvEnrich` returns a sentinel
+// "directive" context (what buildContext now produces); mergeInfoWithOsvContext
+// stays real so the non-osv preserve path still behaves exactly as in production.
+const OSV_DIRECTIVE_SENTINEL = "### Dependency reachability (required) — DIRECTIVE SENTINEL";
+vi.mock("../setup/osv-enrich.js", async (importActual) => {
+  const actual = await importActual<typeof import("../setup/osv-enrich.js")>();
+  return {
+    ...actual,
+    runOsvEnrich: vi.fn(() => ({ context: OSV_DIRECTIVE_SENTINEL, hits: [] })),
+  };
+});
+
 const originalCwd = process.cwd();
 afterEach(() => process.chdir(originalCwd));
 
@@ -245,6 +258,27 @@ describe("one-shot setup coordinator", () => {
     expect(fs.readFileSync(path.join(workspace, "data", "app", "INFO.md"), "utf8")).toBe(
       curatedInfo,
     );
+
+    // Enabling osv on a project that already has a complete INFO.md must NOT
+    // discard the freshly analyzed reachability verdicts by restoring the old
+    // copy with the raw grading directive appended. The fresh analysis wins.
+    fs.writeFileSync(path.join(workspace, "data", "app", "INFO.md"), curatedInfo);
+    const infoAfterOsv = await (async () => {
+      await runSetupWorkflow({
+        workspaceDir: workspace,
+        projectId: "app",
+        projectRoot: project,
+        through: "threat-model",
+        osv: true,
+        services,
+        onLog: () => undefined,
+      });
+      return fs.readFileSync(path.join(workspace, "data", "app", "INFO.md"), "utf8");
+    })();
+    expect(analyze).toHaveBeenCalledTimes(3); // osv flag invalidated the info checkpoint
+    expect(infoAfterOsv).not.toContain(OSV_DIRECTIVE_SENTINEL); // directive never persisted
+    expect(infoAfterOsv).not.toBe(curatedInfo); // old copy not restored over fresh analysis
+    expect(infoAfterOsv).toContain("## Threat model"); // fresh analysis content kept
   });
 
   it("uses the new harness default model instead of a previous harness model", async () => {
diff --git a/packages/deepsec/src/setup/coordinator.ts b/packages/deepsec/src/setup/coordinator.ts
index 54f551f..a7555f1 100644
--- a/packages/deepsec/src/setup/coordinator.ts
+++ b/packages/deepsec/src/setup/coordinator.ts
@@ -551,7 +551,16 @@ export async function runSetupWorkflow(
       // appended section first, then append the freshly generated one (or none).
       // Otherwise refreshes accumulate duplicate sections and keep listing
       // dependencies that have since been patched or removed.
-      if (preserveInfo) {
+      //
+      // Exception: when osv is enabled, `osvEnrichment.context` is a *prompt
+      // directive* (it tells the analysis agent to emit a graded Dependency
+      // reachability block), not a self-contained section. Appending it to the old
+      // INFO.md would persist raw grading instructions and discard the real
+      // reachability verdicts the agent just wove into the freshly generated
+      // INFO.md. Enabling osv is a deliberate "re-analyze with dependencies in
+      // view" action, so keep the fresh analysis (verdicts included) rather than
+      // the preserved copy. The non-osv path keeps curated content as before.
+      if (preserveInfo && !osvEnabled) {
         atomicWriteFileSync(infoPath, mergeInfoWithOsvContext(preserveInfo, osvEnrichment.context));
       }
       state.infoDigest = digest(fs.readFileSync(infoPath, "utf8"));