Back to plugin

Security audit

sil

Security checks for vulnerabilities and agentic risk

Overview

ClawScan rates this plugin suspicious because it found that a gateway method marked for read access can also delete data from the SIL account signed in on that gateway.

Consider the reported read-to-delete privilege mismatch before installing. The scan says a paired gateway client with read access could potentially delete rows from the SIL account signed in on that gateway; it did not dynamically test whether the host or backend permits the described attack. Install only if you are comfortable with that risk.

Vulnerability Patterns
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
  • Agent Memory PoisoningWrites attacker-controlled rules into memory that affect later sessions
  • Remote Payload Retrieval and ExecutionFetches external code whose behavior can change after review
  • Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
Findings (1)

MCP07

Warning
Location
dist/gateway/screen.js:45
Finding

Read-Scoped Gateway Method Allows Deletion of SIL Account Data ## Remediation Suggestions 1. **Separate read and destructive gateway operations.** Remove `shopping_user_forget` from the read-scoped dispatcher and expose deletion through a separate method requiring the host's write scope, such as `operator.write`. 2. **Enforce authorization server-side.** If a shared dispatcher is retained, verify the authenticated caller's write authorization before dispatching any mutation. Do not rely on tool descriptions, model behavior, or UI confirmation. 3. **Bind destructive requests to the intended account.** Verify that the authorized caller is allowed to act on the locally signed-in SIL account. Reject the operation if the account changes between authorization and dispatch. 4. **Validate deletion arguments against the committed schema.** Enforce nonempty string identifiers, identifier length limits, the 50-item maximum, and rejection of additional properties. Treat validation as defense in depth, not a replacement for authorization. 5. **Add authorization regression tests.** Verify that: - A read-only caller can perform intended reads but cannot delete rows. - A direct WebSocket request cannot bypass deletion authorization. - An appropriately write-authorized caller can delete only valid, explicitly named rows. - Read-scoped calls cannot use the `all: true` form. - Account changes invalidate pending destructive actions. 6. **Record destructive actions safely.** Audit the authenticated gateway actor, target account, operation, and outcome without logging tokens or sensitive profile contents.

Content
View full analysis

Vulnerability Details

File Location:

  • plugin/dist/gateway/screen.js:22–30, 45–52, 114–148
  • plugin/dist/tools/catalog.js:37–42
  • plugin/dist/lib/shopping-call.js:23–28

Vulnerability Type: Improper Authorization / Read-to-Write Privilege Escalation (CWE-862)

Risk Level: Medium

Vulnerable Code

The gateway permits the destructive tool alongside read operations and registers the shared dispatcher with read-only scope:

js
const SCREEN_TOOLS = new Set([
    "shopping_product_get",
    "shopping_offers",
    "shopping_seller_get",
    "shopping_domain_get",
    "shopping_content",
    "shopping_user_read",
    "shopping_user_forget",
]);
js
export function registerScreenMethods(api) {
    const openclawVersion = readHostVersion(api);
    const run = {
        pluginVersion: readInstalledVersion(),
        ...(openclawVersion !== null ? { openclawVersion } : {}),
    };
    api.registerGatewayMethod(ANSWER_METHOD, ({ params, respond }) => guarded(api, respond, () => answer(api, params, respond)), { scope: "operator.read" });
    api.registerGatewayMethod(CALL_METHOD, ({ params, respond }) => guarded(api, respond, () => call(api, run, params, respond)), { scope: "operator.read" });
}

The parser restricts deletion to named rows but performs no additional authorization check:

js
function parseCall(params) {
    const { tool: name, args, requestId, sessionId } = params;
    const tool = SHOPPING_TOOLS.find((t) => t.name === name && SCREEN_TOOLS.has(t.name));
    if (tool === undefined)
        return { invalid: `sil.call runs only ${[...SCREEN_TOOLS].join(", ")}.` };
    if (args === null || typeof args !== "object" || Array.isArray(args)) {
        return { invalid: "sil.call takes `args`, the tool's arguments as an object." };
    }
    if (typeof requestId !== "string" || !REQUEST_ID.test(requestId)) {
        return { invalid: "sil.call takes `requestId`, 1–256 printable characters, no spaces." };
    }
    if (sessionId !
...[truncated 4602 chars]

SkillSpector findings are pending for this release.

Endor dependency reachability

Checked Nov 11, 2026 · 8:44 AM UTC

No reachable vulnerabilities reported

Endor dependency scan completed.