Back to skill

Security audit

Thai Chinese Gov Efficiency

Security checks for vulnerabilities and agentic risk

Overview

The skill is mostly a research workflow guide, but it instructs weekly unpinned updates to another skill, which can change future agent behavior without clear user approval.

Review this skill before installing if you do not want an agent to update other skills automatically. Only use it with explicit approval gates for `clawhub update`, preferably pinned versions or reviewed updates, and check any local scripts it is asked to run.

Vulnerability Patterns
  • 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
  • Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
Findings (1)

T08 · Insecure Dependencies

Warning
Location
SKILL.md:71
Finding

Unpinned Automatic Skill Update Creates a Mutable Supply-Chain Risk

Content
View full analysis

Vulnerability Details

File Location: SKILL.md, line 71
Vulnerability Type: T08: Insecure Dependencies
Risk Level: Medium

Vulnerable Code

bash
clawhub update --skill academic-research

The surrounding instruction requires this update to run weekly to synchronize the latest research methodology.

Technical Analysis

The Skill instructs the agent to periodically update the academic-research Skill without specifying an immutable version, trusted commit, integrity hash, signature requirement, change-review process, or approval gate. Consequently, the effective dependency can change after this package has been audited.

If the upstream Skill, publisher account, distribution service, or update channel is compromised, a later update could introduce unauthorized instructions or executable behavior. The reviewed project does not itself contain a malicious payload, and exploitation depends on the behavior of the external update mechanism and subsequently retrieved content.

Attack Path

  1. An attacker compromises the upstream academic-research Skill, its publisher account, or the distribution channel.
  2. The attacker publishes a modified release containing malicious instructions or references to malicious executable content.
  3. The weekly instruction invokes clawhub update --skill academic-research.
  4. Because no immutable version, integrity verification, or manual approval is required, the altered release is installed or synchronized.
  5. A later agent session loads or invokes the updated Skill.
  6. The malicious content then operates with the tools and permissions available to that agent session.

Impact Assessment

Successful exploitation could modify future agent behavior and potentially cause unauthorized tool calls, data access, network requests, or command execution if those capabilities are granted to the updated Skill. The attainable privileges are limited to those of the accou ...[truncated 265 chars]

Remediation
View remediation

Remediation Suggestions

  1. Replace automatic latest-version updates with an explicitly pinned, audited version or immutable commit identifier.
  2. Verify downloaded artifacts using a trusted cryptographic signature and a pinned checksum before installation.
  3. Restrict updates to an authenticated, allowlisted registry and verify the publisher identity.
  4. Download updates into a staging environment rather than activating them immediately.
  5. Review instruction, script, permission, and dependency changes before promotion to production.
  6. Require explicit human approval for every update and record the approver, version, checksum, and review result.
  7. Run updated Skills with least privilege, limiting filesystem, network, credential, and command-execution access.
  8. Maintain a known-good version and a documented rollback procedure.
Vulnerability Patterns
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
Findings (1)

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

This markdown skill mixes an English manifest description with Chinese-only operational guidance and headings for the actual skill behavior. That effectively forces a specific language for users consuming the instructions, without documenting a user choice or justified locale restriction.

Content

No source excerpt is available for this finding.