add-fast-text

Pass

Audited by Gen Agent Trust Hub on Sep 7, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and follow integrationNotes provided by the threenix MCP tool. This creates a surface for indirect prompt injection, where malicious instructions embedded in the tool's response could be executed by the agent.
  • Ingestion points: Output from the get_component_reference tool (SKILL.md, Step 3).
  • Boundary markers: None identified; the agent is instructed to follow the notes verbatim.
  • Capability inventory: The skill has the ability to write files, install package dependencies, and execute build/type-check commands.
  • Sanitization: No sanitization or validation of the tool's output is performed before integration.
  • [DYNAMIC_EXECUTION]: The skill dynamically writes source code to the project directory and decodes base64 content fetched from a remote tool (SKILL.md, Step 4).
  • [EXTERNAL_DOWNLOADS]: The skill installs software dependencies returned by the threenix MCP tool, which could include unverified third-party packages (SKILL.md, Step 6).
  • [COMMAND_EXECUTION]: The skill executes build and type-checking commands on the modified codebase, which could trigger execution of injected code (SKILL.md, Step 7).
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 7, 2026, 08:11 AM
Security Audit — agent-trust-hub — add-fast-text