day2-supplement-mcp

Fail

Audited by Gen Agent Trust Hub on Mar 6, 2026

Risk Level: HIGHREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
  • [REMOTE_CODE_EXECUTION]: The skill uses npx skills add ai-native-camp/camp-1 --agent claude-code --yes to download and install additional agent components. This executes remote code/packages from an unverified source without user confirmation due to the --yes flag.
  • [COMMAND_EXECUTION]: The skill triggers shell commands (npx) to modify the agent's skill set, effectively changing its operational capabilities at runtime.
  • [EXTERNAL_DOWNLOADS]: The skill references and downloads content from a GitHub repository (ai-native-camp/camp-1) that is not part of the trusted vendors list or well-known service organizations.
  • [PROMPT_INJECTION]: The skill contains a 'STOP PROTOCOL' that uses aggressive override language ('Must never violate', 'Top priority rule', 'Absolutely forbidden') to force the agent to ignore standard interaction patterns and adhere to a specific, author-defined execution flow.
  • [PROMPT_INJECTION]: There is a significant surface for indirect prompt injection as the skill reads content from multiple external files in the references/ directory and displays them to the user or uses them to generate responses without sanitization.
  • Ingestion points: Multiple files including references/block0-concept.md and references/block1-add-server.md.
  • Boundary markers: None identified; content is used as-is.
  • Capability inventory: The agent has access to shell execution (bash) and interactive tool calling (AskUserQuestion).
  • Sanitization: No evidence of validation or sanitization of the external markdown content before processing.
Recommendations
  • AI detected serious security threats
Audit Metadata
Risk Level
HIGH
Analyzed
Mar 6, 2026, 03:23 PM