skill-share

Fail

Audited by Gen Agent Trust Hub on Feb 16, 2026

Risk Level: HIGHPROMPT_INJECTIONCOMMAND_EXECUTIONDATA_EXFILTRATION
Full Analysis
  • [Indirect Prompt Injection] (HIGH): The skill creates an attack surface by ingesting untrusted user input (skill name and description) and using it to generate SKILL.md files, directory structures, and Slack notifications.
  • Ingestion points: User-provided metadata during the 'Initialization' step in SKILL.md.
  • Boundary markers: None identified in the provided documentation to separate user input from generated instruction/code templates.
  • Capability inventory: Write access to the filesystem (creating directories/zips) and network access via Slack integration (Rube).
  • Sanitization: No sanitization logic is described for the user input used in script generation or Slack block formatting.
  • [Dynamic Execution] (MEDIUM): The skill documentation explicitly mentions it 'Generates standardized scripts/' and 'Auto-generates YAML frontmatter'. Writing executable scripts to the local filesystem based on user-provided metadata is a form of dynamic code generation that could be exploited to embed malicious logic into the 'created' skills.
  • [Data Exfiltration] (MEDIUM): The skill leverages Rube for Slack integration (SLACK_SEND_MESSAGE, SLACK_POST_MESSAGE_WITH_BLOCKS). While intended for discovery, this mechanism could be abused to exfiltrate sensitive data if an injection attack influences the 'summary' or 'metadata' sent to the Slack workspace.
Recommendations
  • AI detected serious security threats
Audit Metadata
Risk Level
HIGH
Analyzed
Feb 16, 2026, 03:41 AM