ln-81-skill-reviewer

Fail

Audited by Gen Agent Trust Hub on Oct 18, 2026

Risk Level: HIGHREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
  • [REMOTE_CODE_EXECUTION]: The instructions require the agent to 'Discover and run every repository-required skill validator' and 'Run every repository-required plugin or package validator' found in the target repository. Because these scripts are defined by the repository being reviewed (which may be untrusted), this facilitates the execution of arbitrary code from remote sources.
  • [COMMAND_EXECUTION]: The skill uses shell interaction as a primary mechanism, including 'Equivalent read-only shell commands' and the execution of various native and repository-defined validators. This provides a direct path for malicious commands to be executed if triggered by external content.
  • [DYNAMIC_EXECUTION]: The workflow involves identifying and running executable content (validators, tools) that is not part of the skill's static definition but is discovered at runtime within the data it processes. This dynamic loading and execution of untrusted logic is a high-risk pattern.
  • [INDIRECT_PROMPT_INJECTION]: This category identifies a high attack surface where the skill ingests untrusted data and has extensive capabilities.
  • Ingestion points: The agent is instructed to read 'repository instructions, skill contracts, and every configured host catalog' (SKILL.md) which are provided by the external repository under review.
  • Boundary markers: The skill attempts to mitigate risks through instructions like 'Cite each finding to an exact file' and 'distinguish facts, inferences, and unverified claims', but these do not prevent adversarial injection in the processed data.
  • Capability inventory: The agent possesses capabilities for 'native file search', shell command execution, and web research ('Search and read primary pages for current external recommendations').
  • Sanitization: There are no technical sanitization or validation steps for the external content before it is processed by the agent; the skill relies purely on instructions to 'read selectively'.
Recommendations
  • AI detected serious security threats
Audit Metadata
Risk Level
HIGH
Analyzed
Oct 18, 2026, 05:49 PM