Back to skill

Security audit

Google Super

Security checks for vulnerabilities and agentic risk

Overview

This skill is a disclosed Google integration, but it asks users to install and allowlist an unpinned third-party plugin and connect broad Google access across many sensitive services through one OAuth flow.

Review this carefully before installing. Use a dedicated or low-risk Google account where possible, avoid connecting production email, Drive, Ads, or Analytics accounts unless you need the full bundle, confirm sensitive reads and all writes yourself, and revoke the ClawLink Google OAuth grant when finished.

Vulnerability Patterns
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
  • Insecure DependenciesIntroduces malicious components through unsafe dependency sources
  • 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
Findings (2)

T05 · Unauthorized Access and Privilege Escalation

Error
Location
SKILL.md:53
Finding
Overly Broad Cross-Service Google OAuth Authorization## Vulnerability Details **File Location**: `SKILL.md:53` **Vulnerability Type**: Excessive OAuth permissions and failure to enforce least privilege **Risk Level**: High ### Vulnerable Code Snippet ```markdown ## Authentication ClawLink manages the full Google OAuth flow with broad scopes. Connect at [claw-link.dev/dashboard?add=google-super](https://claw-link.dev/dashboard?add=google-super). The connection grants access to all bundled Google services. ``` Related security guidance at `SKILL.md:65-69`: ```markdown ## Security & Permissions - **Read** tools are safe and require no confirmation - **Write** tools require confirmation before execution - Batch delete, clear values, and data erasure are high-impact - Access is scoped by the Google OAuth permissions granted during connection ``` ### Technical Analysis The Skill directs users to grant a hosted third-party OAuth intermediary broad authorization covering all bundled Google services. The declared integration includes Gmail, Calendar, Drive, Docs, Sheets, Slides, Meet, Tasks, Photos, Maps, Google Analytics, and Google Ads, with more than 400 tools available through one connection. This authorization model does not follow least privilege for common requests. A user who only needs to send an email, list calendar events, or update one spreadsheet is instructed to authorize unrelated services and data categories. The assertion that all read operations are safe and require no confirmation is also unsafe. Read operations can retrieve sensitive email content, Drive documents, meeting transcripts, recordings, photos, account metadata, analytics information, and advertising data. Although the network transmission through ClawLink is part of the declared hosted integration rather than a concealed exfiltration mechanism, it creates a third-party trust boundary for highly sensitive information. ### Attack Path 1. A user installs the integration and connects a ...[truncated 1461 chars]
Remediation
## Remediation Suggestions 1. Replace the single broad authorization flow with incremental, service-specific OAuth authorization. 2. Request only the scopes needed for the operation explicitly selected by the user. 3. Separate unrelated services into independently enabled integrations or permission groups. 4. Default all unrelated tools and destructive capabilities to disabled. 5. Require explicit, context-rich confirmation for sensitive reads as well as writes. The confirmation should identify the service, resource, fields being accessed, recipient or destination, and expected effect. 6. Require stronger confirmation for permanent deletion, bulk mutation, file sharing, email transmission, and advertising-account changes. 7. Provide an interface that shows currently granted scopes and allows users to revoke individual services without disconnecting the entire integration. 8. Minimize data forwarded through ClawLink, encrypt data in transit and at rest, apply short retention periods, and document subprocessors and logging behavior. 9. Enforce server-side authorization per tool rather than relying only on catalog visibility or agent instructions. 10. Replace the statement that all read tools are safe with a risk-based policy that treats private content retrieval as a sensitive operation.

T08 · Insecure Dependencies

Warning
Location
SKILL.md:33
Finding
Unpinned Third-Party Plugin Installation and Allowlisting## Vulnerability Details **File Location**: `SKILL.md:33-35` **Vulnerability Type**: Mutable third-party dependency installed without a version or integrity constraint **Risk Level**: Medium ### Vulnerable Code Snippet ```bash openclaw plugins install clawhub:clawlink-plugin openclaw config set tools.alsoAllow '["clawlink-plugin"]' --strict-json openclaw gateway restart ``` ### Technical Analysis The installation instructions retrieve `clawhub:clawlink-plugin` without pinning a reviewed version or cryptographic digest. They then add the plugin to the allowed-tools configuration and restart the gateway. Because the plugin implementation is not contained in the audited project, its effective behavior cannot be verified from `SKILL.md`. In particular, the audit cannot validate its OAuth-token handling, network destinations, confirmation enforcement, data retention, or implementation of the advertised tools. An unpinned package reference can resolve to code that changes after the Skill has been reviewed. This introduces a supply-chain trust boundary: compromise of the package source, publishing account, registry, or a future release could change the runtime behavior while the documentation and reviewed Skill file remain unchanged. The available evidence does not establish that the current plugin is malicious. The vulnerability is the absence of version and integrity controls for a highly privileged external dependency. ### Attack Path 1. An attacker compromises the plugin publisher, package source, distribution registry, or a future plugin release. 2. The attacker publishes a modified package under the same unversioned plugin identifier. 3. A user follows the documented installation command, which resolves to the modified package. 4. The user explicitly allowlists the plugin and restarts the OpenClaw gateway. 5. The modified plugin executes within the enabled tool environment. 6. Once the user connects Google, the p ...[truncated 808 chars]
Remediation
## Remediation Suggestions 1. Pin the plugin to an immutable, reviewed version rather than installing by an unconstrained package name. 2. Verify the package with a cryptographic digest or signed provenance record before installation. 3. Publish the plugin source, build instructions, permission manifest, and reproducible build artifacts so its behavior can be independently audited. 4. Use a trusted registry namespace with protected publishing credentials and mandatory multi-factor authentication. 5. Enforce package-signature verification and reject unsigned or unexpectedly changed releases. 6. Restrict the plugin to explicitly approved network hosts and tools. 7. Isolate the plugin using sandboxing and prevent unnecessary access to local files, environment variables, credentials, and unrelated tools. 8. Couple each plugin release to a reviewed OAuth scope manifest and block silent scope expansion. 9. Document a rollback and revocation process for compromised releases, including OAuth-token revocation. 10. Perform a separate source-code audit of the actual plugin before granting it access to production Google accounts.
Vulnerability Patterns
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Findings (3)

Vague Triggers

Medium
Confidence
93% confidence
Finding
The description is extremely broad and overlaps with common user intents like email, calendar, documents, search, and analytics, increasing the chance the agent invokes this skill by default even when narrower or safer tools would suffice. Because the integration grants cross-service Google access through one connection, overbroad routing can expose far more user data than is necessary for the task.

Missing User Warnings

Medium
Confidence
90% confidence
Finding
The skill states that OAuth grants access to all bundled Google services with broad scopes, but it does not prominently warn users that connecting this integration may expose email, files, calendars, photos, analytics, and ads data through a single authorization. This omission undermines informed consent and can lead users or downstream agents to authorize a much wider data-sharing surface than they expect.

Autonomous Decision Making

Medium
Category
Excessive Agency
Content
## Security & Permissions

- **Read** tools are safe and require no confirmation
- **Write** tools require confirmation before execution
- Batch delete, clear values, and data erasure are high-impact
- Access is scoped by the Google OAuth permissions granted during connection
Confidence
84% confidence
Finding
Labeling all read tools as 'safe' and requiring 'no confirmation' is overly permissive for a multi-service integration that can read highly sensitive content such as Gmail, Drive documents, Meet transcripts, Photos, and analytics data. In an agent setting, this can normalize autonomous access to confidential information without user awareness, especially because the skill aggregates hundreds of tools under one OAuth grant.

Static analysis

No suspicious patterns detected.