strategy-dev-manager

Warn

Audited by Gen Agent Trust Hub on Oct 30, 2026

Risk Level: MEDIUMINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTIONPERSISTENCECOMMAND_EXECUTION
Full Analysis
  • [DYNAMIC_EXECUTION]: The skill generates Python code (signal_engine.py) using provided templates and user-extracted logic from papers, which is subsequently executed via the backtest tool. This introduces a risk if the generation logic is influenced by malicious content.
  • [INDIRECT_PROMPT_INJECTION]: The skill has a large attack surface for indirect prompt injection due to its core workflow:
  • Ingestion points: Untrusted data enters the agent context through the read_document tool, which parses external PDF papers and research reports (SKILL.md Phase 1).
  • Boundary markers: The instructions do not specify any delimiters or safety warnings to prevent the agent from following instructions embedded within the paper text.
  • Capability inventory: The skill possesses high-impact capabilities, including the ability to generate and execute Python code via backtest and perform syntax checks using bash (SKILL.md Phase 3).
  • Sanitization: There is no documented sanitization or validation of the extracted mathematical formulas or logic before they are used to scaffold and implement the executable SignalEngine.
  • [PERSISTENCE]: The skill utilizes a ScheduledResearchExecutor (documented in references/scheduled_decay_scan.md) which allows for the creation of persistent, scheduled jobs that fire research prompts periodically. This background execution mechanism could be leveraged to maintain long-term persistence for malicious tasks or exfiltration loops.
  • [COMMAND_EXECUTION]: In Phase 3 (IMPLEMENT), the skill executes a shell command via bash("python -c ...") to validate the syntax of the generated signal_engine.py script. While intended for validation, this represents a direct command execution path for locally generated content.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Oct 30, 2026, 12:04 PM