Back to skill

Security audit

Medici Investments Pack

Security checks for vulnerabilities and agentic risk

Overview

This is a small documentation-only skill pack that tells users to install two named trading-related skills, with a supply-chain caution because those installs are not pinned to reviewed versions.

Before installing, confirm that the two referenced Medici Investments skills are the intended packages from a trusted publisher and, where possible, install reviewed or pinned versions. The pack itself is documentation-only, but the installed child skills may have their own permissions and behavior.

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:25
Finding
Unpinned External Skill Installation Without Integrity Verification## Vulnerability Details **File Location**: `SKILL.md`, lines 25–28 **Vulnerability Type**: Supply-chain risk from mutable, externally distributed skill dependencies **Risk Level**: Medium ```bash ```bash clawhub install medici-investments-position-sizer-dv clawhub install medici-investments-pulse-dv ``` ``` ### Technical Analysis The documented installation commands retrieve two external skills by package name without specifying an immutable version, content hash, verified source repository, or publisher identity. Neither external skill is included in the audited project, so its implementation and effective instructions cannot be reviewed from the supplied files. Because package names can resolve to content that changes after this project has been audited, users may install code or agent instructions different from those originally intended. A compromised registry account, namespace takeover, malicious package update, or repository compromise could therefore turn either dependency into a supply-chain delivery channel. This finding does not establish that the named packages are currently malicious. It identifies the absence of controls needed to ensure that future installations retrieve the same reviewed content. ### Attack Path 1. An attacker compromises the publisher account, registry infrastructure, or release process for one of the named external skills, or otherwise gains control of the corresponding package namespace. 2. The attacker publishes a modified release containing malicious scripts or agent instructions. 3. A user follows the Quick Start commands in `SKILL.md`. 4. ClawHub resolves the unpinned package name to the attacker-controlled release. 5. The downloaded skill is loaded or invoked with the permissions available to the user's agent environment. 6. The malicious release can perform actions permitted by that environment, potentially including accessing agent-visible data, invoking enabled tools, mo ...[truncated 701 chars]
Remediation
## Remediation Suggestions 1. Pin each external skill to a reviewed, immutable version rather than installing the latest release by package name alone. 2. Where supported, verify downloads using cryptographic content hashes or signed release metadata. 3. Document the expected publisher identity and canonical source repository for each dependency. 4. Vendor the reviewed skill contents into the project when licensing and maintenance constraints permit. 5. Add a review step that inspects downloaded scripts, manifests, and instructions before activation. 6. Run third-party skills with least privilege, restricting filesystem access, credentials, enabled tools, and outbound network connectivity. 7. Use dependency monitoring and an approved-version policy so updates are not adopted automatically without security review.
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

Static analysis

No suspicious patterns detected.