ghidra-headless-auto-evolution

Warn

Audited by Gen Agent Trust Hub on Apr 1, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill is authorized to "directly create or update" executable and configuration files, including skill files, reusable scripts, and new child-skill entry points within the .agents/skills/ directory. This grants the agent the ability to persistently modify its own behavioral instructions and toolset based on its execution history.
  • [PROMPT_INJECTION]: The skill presents an indirect prompt injection surface while processing and generalizing previous task artifacts.
  • Ingestion points: Artifacts, notes, scripts, and evidence records from completed tasks, as specified in the "Required Inputs" section of SKILL.md.
  • Boundary markers: The skill includes checkpoints for task context and benefit statements but lacks explicit delimiters or "ignore previous instructions" warnings when processing the untrusted content from reviewed artifacts.
  • Capability inventory: The agent has the authority to write to the .agents/skills/ directory, affecting its own system instructions and logic for future sessions.
  • Sanitization: No explicit sanitization, validation, or escaping protocols are defined for the content extracted from artifacts before it is written into permanent tracked assets.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Apr 1, 2026, 05:56 AM