brainstorm-with-file

Pass

Audited by Gen Agent Trust Hub on Mar 2, 2026

Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill executes a local shell command to identify the environment context.
  • Evidence: The script runs git rev-parse --show-toplevel to determine the project root directory for saving workflow artifacts.
  • [PROMPT_INJECTION]: User-supplied data is interpolated into instructions for subagents, creating a surface for indirect prompt injection.
  • Ingestion points: The TOPIC argument provided by the user and the codebase context gathered via grep/glob in Step 2.1 are both used to build subagent prompts.
  • Boundary markers: The skill uses Markdown headers (e.g., ## Task, ## Context) to separate instructions from data, but lacks explicit "ignore embedded instructions" guards.
  • Capability inventory: The skill can read local files, write JSON/Markdown artifacts, and spawn additional agents with new instructions.
  • Sanitization: No sanitization or escaping is performed on the ${idea_or_topic} or ${explorationContext} variables before they are inserted into the JavaScript templates for spawn_agent calls.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 2, 2026, 08:23 PM