agenticflow-skills

Pass

Audited by Gen Agent Trust Hub on Feb 26, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [SAFE]: The skill consists entirely of reference documentation and does not contain any executable scripts, binaries, or hidden malicious logic.\n- [EXTERNAL_DOWNLOADS]: Documents the installation of the @pixelml/agenticflow-cli package. This is a vendor-owned tool (pixelml) and is essential for interacting with the AgenticFlow platform as described.\n- [COMMAND_EXECUTION]: Provides guidance for a code_executor tool within the platform that allows for running Python, JavaScript, and Bash. The documentation specifies that these functions are intended to run in a sandboxed environment.\n- [PROMPT_INJECTION]: Identifies the attack surface for indirect prompt injection within the documented platform's architecture for handling user-defined inputs.\n
  • Ingestion points: Untrusted data enters the system through input_schema fields documented in reference/workflow/overview.md.\n
  • Boundary markers: The platform documentation defines "Guardrails" to constrain agent behavior and persona as noted in the glossary and agent overview.\n
  • Capability inventory: The platform supports tools with significant capabilities including http_request, file_operations, and code_executor.\n
  • Sanitization: Documentation provides high-level best practices for security, such as using minimal tools and explicit guardrails, but does not provide specific implementation logic for sanitizing all inputs.
Audit Metadata
Risk Level
SAFE
Analyzed
Feb 26, 2026, 11:57 AM