idea-creator

Warn

Audited by Gen Agent Trust Hub on Nov 10, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTIONDATA_EXFILTRATION
Full Analysis
  • [COMMAND_EXECUTION]: The skill utilizes Bash logic to resolve tool paths (e.g., research_wiki.py, threat_scan.py, and reviewer_route.py) by reading local files such as .aris/installed-skills-codex.txt and $HOME/.aris/repo. These tools are then executed via python3 with arguments derived at runtime.
  • [DYNAMIC_EXECUTION]: The skill uses a fallback mechanism to write files via cat and executes Python scripts whose locations are determined dynamically from the user's home directory and project state.
  • [INDIRECT_PROMPT_INJECTION]: The skill ingests research-wiki/query_pack.md as untrusted input to drive research idea generation. While it uses a threat scanner, the raw content is processed to define search seeds, research gaps, and banlists.
  • Ingestion points: research-wiki/query_pack.md.
  • Boundary markers: Uses an external threat_scan.py for validation but lacks internal delimiters or safety prompts for the ingested content.
  • Capability inventory: WebSearch, spawn_agent, file writing, and local script execution.
  • Sanitization: Relies on external validation with no explicit filtering.
  • [DATA_EXFILTRATION]: The skill accesses local configuration files in hidden directories, such as .aris/installed-skills-codex.txt and $HOME/.aris/repo, to resolve tool paths.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Nov 10, 2026, 09:37 AM