strategy-generate

Pass

Audited by Gen Agent Trust Hub on Nov 21, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTIONCOMMAND_EXECUTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user descriptions to derive strategy logic and generate Python code. This ingestion of external data without proper boundary markers or specific sanitization creates a surface for prompt injection attacks where malicious instructions could be embedded in the user's strategy description.
  • Ingestion points: User strategy requirements parsed from the natural language descriptions in SKILL.md.
  • Boundary markers: Absent; there is no explicit separation between untrusted user input and the agent's internal prompt instructions.
  • Capability inventory: Capabilities include file system writes (write_file), shell command execution (bash), and backtest execution (backtest) that runs the generated code.
  • Sanitization: No explicit sanitization or filtering of the user-provided description is implemented before the agent uses it to generate executable code.
  • [DYNAMIC_EXECUTION]: The skill's core workflow involves generating a Python script (code/signal_engine.py) at runtime based on user requirements and subsequently executing it. This dynamic generation and execution from untrusted input is a high-risk pattern, although it is intrinsic to the skill's purpose as a developer tool.
  • [COMMAND_EXECUTION]: The skill uses a bash tool to perform a Python syntax check on the generated file. While the specific command (ast.parse) is a safe static analysis method, the use of shell execution to process dynamically generated content is a sensitive operation.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 21, 2026, 05:36 PM