create-profile-atlas

Warn

Audited by Gen Agent Trust Hub on Jul 7, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The Python script 'scripts/profile_atlas_postprocess.py' is vulnerable to path traversal. It constructs file paths for output images using the user-supplied 'name' variable (e.g., 'out_dir / f"{name}.png"') without sanitization. This allows a malicious user to provide names containing path traversal sequences to write files to locations outside the intended output directory.
  • [PROMPT_INJECTION]: The instructions in 'SKILL.md' explicitly tell the agent to use the '--yes' flag with 'agent-slack' commands for adding and removing emoji. This bypasses human-in-the-loop confirmation for destructive actions in the Slack workspace, increasing the risk of unauthorized modifications.
  • [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted input that influences file system and API operations.
  • Ingestion points: User-provided shortcode names and prefixes in 'SKILL.md'.
  • Boundary markers: Absent.
  • Capability inventory: Local file system write access via 'scripts/profile_atlas_postprocess.py' and Slack management via 'agent-slack'.
  • Sanitization: Absent; the script does not validate filename components.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Jul 7, 2026, 09:40 AM
Security Audit — agent-trust-hub — create-profile-atlas