executing-plans

Warn

Audited by Gen Agent Trust Hub on Feb 16, 2026

Risk Level: MEDIUMPROMPT_INJECTION
Full Analysis
  • [PROMPT_INJECTION] (MEDIUM): The skill is vulnerable to Indirect Prompt Injection because it is designed to ingest and execute instructions from an external, untrusted 'plan file'.
  • Ingestion points: Step 1 requires the agent to read an external file containing an implementation plan.
  • Boundary markers: No boundary markers or specific delimiters are used to separate the plan content from the agent's system instructions.
  • Capability inventory: While not explicitly defined in the Markdown, the 'execute tasks' and 'implement' descriptions imply capabilities like file modification and command execution.
  • Sanitization: No technical sanitization of the plan content is performed.
  • Instructional Risk: The directive 'Follow each step exactly' in Step 2 increases the likelihood of executing malicious embedded instructions. However, this is partially mitigated by the requirement to 'Review critically' and the mandatory human-in-the-loop checkpoints ('Ready for feedback').
Audit Metadata
Risk Level
MEDIUM
Analyzed
Feb 16, 2026, 01:02 PM