explore-code

Warn

Audited by Gen Agent Trust Hub on May 10, 2026

Risk Level: MEDIUMREMOTE_CODE_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [REMOTE_CODE_EXECUTION]: The script scripts/write_outputs.py utilizes importlib.util to dynamically load and execute a module from a path computed at runtime (../../../shared/scripts/write_explore_bundle.py). This technique allows for the execution of code residing outside the skill's controlled package environment, which is a significant security risk as it bypasses standard module isolation.
  • [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection via scripts/plan_code_changes.py. It reads untrusted JSON data from user-provided file paths and uses this content to construct instructions and adaptation tracks that guide the agent's behavior.
  • Ingestion points: Untrusted data is ingested from files specified by the --variant-spec-json, --idea-card-json, and --analysis-json command-line arguments.
  • Boundary markers: There are no boundary markers or instructions to ignore embedded commands within the processed JSON data.
  • Capability inventory: The skill includes capabilities for comprehensive file system scanning (rglob) and dynamic code loading/execution (importlib).
  • Sanitization: The skill parses JSON but performs no validation or sanitization against a strict schema to prevent malicious instruction embedding within the research adaptation plan.
Audit Metadata
Risk Level
MEDIUM
Analyzed
May 10, 2026, 12:37 PM