start-research

Fail

Audited by Gen Agent Trust Hub on Feb 23, 2026

Risk Level: HIGHCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION] (HIGH): The file SKILL.md contains a bash script invocation that directly interpolates the user-provided {topic} variable: .claude/hooks/workflows/write-session-state.sh "{topic}" .... Since this variable is gathered directly from user input in references/gather-context.md, an attacker can use shell metacharacters (such as backticks or semicolons) to escape the intended command and execute arbitrary code.
  • [PROMPT_INJECTION] (LOW): The skill is susceptible to indirect prompt injection (Category 8). Evidence: 1. Ingestion points: User input is collected through multiple prompts in references/gather-context.md. 2. Boundary markers: None present. 3. Capability inventory: The skill has the ability to execute bash scripts and write session state. 4. Sanitization: No sanitization or escaping is performed on the gathered user input before it is used in command interpolation or file path construction.
Recommendations
  • AI detected serious security threats
Audit Metadata
Risk Level
HIGH
Analyzed
Feb 23, 2026, 06:03 AM