Skip to content

OpenAI's Codex Security CLI and TypeScript SDK for finding, validating, and fixing security vulnerabilities. npm: https://www.npmjs.com/package/@openai/codex-security

fix(sdk): respect Deep and diff scan completion ownership (#457)

Dane Schneider1fc1a8ce9e49
diff --git a/sdk/typescript/_bundled_plugin/scripts/workbench_db.py b/sdk/typescript/_bundled_plugin/scripts/workbench_db.py
index 1fc4ff72b..c5cb73c34 100644
--- a/sdk/typescript/_bundled_plugin/scripts/workbench_db.py
+++ b/sdk/typescript/_bundled_plugin/scripts/workbench_db.py
@@ -1443,6 +1443,8 @@ def complete_scan_locked(
verify_manifest_binding(scan, manifest)
manifest_digest = published_manifest_digest(scan_dir, manifest)
pin_legacy_manifest_digest(connection, scan["id"], manifest_digest)
+ if cost_json is not None and scan["recipe_json"] is not None:
+ scan_usage.reconcile_completed_scan_cost(connection, scan, cost_json)
return scan_context(connection, scan["id"])
if scan["status"] != "running":
raise SystemExit("Only a running scan can be completed.")
diff --git a/sdk/typescript/_bundled_plugin/scripts/workbench_scan_usage.py b/sdk/typescript/_bundled_plugin/scripts/workbench_scan_usage.py
index be5bc9282..db987d20c 100644
--- a/sdk/typescript/_bundled_plugin/scripts/workbench_scan_usage.py
+++ b/sdk/typescript/_bundled_plugin/scripts/workbench_scan_usage.py
@@ -55,6 +55,32 @@ def measured_scan_cost_json(usage: Mapping[str, Any]) -> str:
return json.dumps({"usage": dict(usage)}, separators=(",", ":"), allow_nan=False)
+def reconcile_completed_scan_cost(
+ connection: sqlite3.Connection,
+ scan: sqlite3.Row,
+ cost_json: str,
+) -> None:
+ """Persist authoritative SDK cost without discarding measured worker usage."""
+
+ existing = json.loads(scan["cost_json"]) if scan["cost_json"] is not None else {}
+ if isinstance(existing, dict) and "usage" in existing:
+ cost_json = json.dumps(
+ {**existing, "cost": json.loads(cost_json)},
+ separators=(",", ":"),
+ allow_nan=False,
+ )
+ connection.execute("BEGIN IMMEDIATE")
+ try:
+ connection.execute(
+ "UPDATE scans SET cost_json = ? WHERE id = ? AND status = 'complete'",
+ (cost_json, scan["id"]),
+ )
+ connection.commit()
+ except BaseException:
+ connection.rollback()
+ raise
+
+
def collect_scan_usage(
connection: sqlite3.Connection,
scan: sqlite3.Row,
diff --git a/sdk/typescript/src/api.ts b/sdk/typescript/src/api.ts
index 3fde8e860..2c56b64bb 100644
--- a/sdk/typescript/src/api.ts
+++ b/sdk/typescript/src/api.ts
@@ -914,6 +914,7 @@ export class CodexSecurity {
runtime.configPath !== undefined,
knowledgeBase !== null,
options.scanPrompt,
+ options.maxCostUsd !== undefined,
);
checkOpen();
const feedback = await workbench(
@@ -2384,6 +2385,7 @@ function scanPrompt(
hasConfigPath = false,
hasKnowledgeBase = false,
additionalPrompt?: string,
+ enforceCostLimit = false,
): string {
const python = `${process.platform === "win32" ? "& " : ""}${shellEnvironmentReference("PYTHON")}`;
return [
@@ -2445,7 +2447,13 @@ function scanPrompt(
: []),
"Runtime paths are environment-backed; keep them quoted in POSIX shells and use the corresponding $env: names in PowerShell. Do not copy or reparse their values.",
targetInstruction(target, python),
- "Write the complete canonical scan-manifest.json, findings.json, and coverage.json, but do not finalize or seal them; the SDK workbench owns authoritative metadata, finalization, report generation, and sealing.",
+ ...(skillName === "security-scan" || enforceCostLimit
+ ? [
+ "Write the complete canonical scan-manifest.json, findings.json, and coverage.json, but do not finalize or seal them; the SDK workbench owns authoritative metadata, finalization, report generation, and sealing.",
+ ]
+ : [
+ "Use record_codex_security_scan_draft and complete_codex_security_scan as directed by the selected skill; the workbench owns authoritative metadata, finalization, report generation, and sealing.",
+ ]),
...(additionalPrompt?.trim()
? ["Additional scan instructions:", additionalPrompt]
: []),
diff --git a/sdk/typescript/tests-ts/api.test.ts b/sdk/typescript/tests-ts/api.test.ts
index a4e0fcdea..2b4847613 100644
--- a/sdk/typescript/tests-ts/api.test.ts
+++ b/sdk/typescript/tests-ts/api.test.ts
@@ -1923,6 +1923,7 @@ describe("CodexSecurity orchestration", () => {
expect(prompt).toContain("$codex-security:security-scan");
expect(prompt).toContain("The SDK has already registered this scan.");
expect(prompt).toContain("never call a scan-start or completion tool");
+ expect(prompt).toContain("do not finalize or seal them");
expect(prompt).toContain(
"This Standard scan authorizes its independent baseline auditor and focused investigators",
);
@@ -5182,6 +5183,8 @@ describe("CodexSecurity orchestration", () => {
`Scan target: Git diff from ${revision} to ${revision}.`,
);
expect(prompt).toContain("$codex-security:security-diff-scan");
+ expect(prompt).toContain("complete_codex_security_scan");
+ expect(prompt).not.toContain("do not finalize or seal them");
expect(prompt).toContain(
"This exhaustive scan authorizes the delegated-worker phases",
);
@@ -5192,6 +5195,8 @@ describe("CodexSecurity orchestration", () => {
"prompt captured",
);
expect(prompt).toContain("$codex-security:deep-security-scan");
+ expect(prompt).toContain("complete_codex_security_scan");
+ expect(prompt).not.toContain("do not finalize or seal them");
expect(prompt).toContain(
'start_codex_security_deep_scan with {"scanId":"scan_example_001"}',
);
@@ -5199,6 +5204,21 @@ describe("CodexSecurity orchestration", () => {
"This exhaustive scan authorizes the delegated-worker phases",
);
+ await expect(
+ client.run(repository, { mode: "deep", maxCostUsd: 1 }),
+ ).rejects.toThrow("prompt captured");
+ expect(prompt).toContain("do not finalize or seal them");
+ expect(prompt).not.toContain("complete_codex_security_scan");
+
+ await expect(
+ client.run(repository, {
+ target: DiffTarget.refs({ base, head }),
+ maxCostUsd: 1,
+ }),
+ ).rejects.toThrow("prompt captured");
+ expect(prompt).toContain("do not finalize or seal them");
+ expect(prompt).not.toContain("complete_codex_security_scan");
+
await expect(
client.run(repository, { target: DiffTarget.workingTree({ base }) }),
).rejects.toThrow("prompt captured");
diff --git a/sdk/typescript/tests-ts/scan-recovery.test.ts b/sdk/typescript/tests-ts/scan-recovery.test.ts
index fa2c7c912..e6709d55d 100644
--- a/sdk/typescript/tests-ts/scan-recovery.test.ts
+++ b/sdk/typescript/tests-ts/scan-recovery.test.ts
@@ -460,6 +460,94 @@ describe("malformed scan artifact recovery", () => {
expect((await completeScan(fixture)).progress.status).toBe("complete");
});
+ test.each([
+ ["unavailable", undefined],
+ [
+ "worker-inclusive",
+ {
+ coverage: "complete",
+ source: "codex_rollout",
+ threadCount: 3,
+ inputTokens: 5_000,
+ cachedInputTokens: 400,
+ cacheWriteInputTokens: 0,
+ outputTokens: 120,
+ reasoningOutputTokens: 20,
+ totalTokens: 5_120,
+ },
+ ],
+ ] as const)(
+ "reconciles authoritative scan cost without replacing %s usage or sealed artifacts",
+ async (_scenario, measuredUsage) => {
+ const fixture = await startDraftScan();
+ const firstCompletion = await workbench(fixture, [
+ "complete-scan",
+ "--scan-id",
+ fixture.scanId,
+ ...(measuredUsage === undefined
+ ? []
+ : ["--cost-json", JSON.stringify({ usage: measuredUsage })]),
+ ]);
+ const initiallyCompleted = firstCompletion["scan"] as ScanSummary & {
+ usage: Record<string, unknown>;
+ cost?: unknown;
+ };
+ expect(initiallyCompleted.progress.status).toBe("complete");
+ expect(initiallyCompleted.usage).toBeDefined();
+ if (measuredUsage !== undefined) {
+ expect(initiallyCompleted.usage).toEqual(measuredUsage);
+ }
+ expect(initiallyCompleted.cost).toBeUndefined();
+
+ const artifactNames = [
+ "scan-manifest.json",
+ "findings.json",
+ "coverage.json",
+ "report.md",
+ ];
+ const sealedArtifacts = await Promise.all(
+ artifactNames.map((name) => readFile(join(fixture.scanDir, name))),
+ );
+ const cost = {
+ model: "gpt-5.6-sol",
+ inputTokens: 1_250,
+ cachedInputTokens: 200,
+ cacheWriteInputTokens: 0,
+ outputTokens: 30,
+ estimatedUsd: 0.00625,
+ };
+
+ const reconciled = await workbench(fixture, [
+ "complete-scan",
+ "--scan-id",
+ fixture.scanId,
+ "--cost-json",
+ JSON.stringify(cost),
+ ]);
+ expect(reconciled["scan"]).toMatchObject({
+ progress: { status: "complete" },
+ usage: initiallyCompleted.usage,
+ cost,
+ });
+
+ const persisted = await workbench(fixture, [
+ "get-scan",
+ "--scan-id",
+ fixture.scanId,
+ ]);
+ expect(persisted["scan"]).toMatchObject({
+ progress: { status: "complete" },
+ usage: initiallyCompleted.usage,
+ cost,
+ });
+ expect(
+ await Promise.all(
+ artifactNames.map((name) => readFile(join(fixture.scanDir, name))),
+ ),
+ ).toEqual(sealedArtifacts);
+ },
+ );
+
test("preserves target-drift classification from prepared completion", async () => {
const fixture = await startDraftScan();
const source = join(fixture.repository, "src", "extract.py");