openspec-writing-plans

Fail

Audited by Gen Agent Trust Hub on Feb 14, 2026

Risk Level: HIGHPROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [Indirect Prompt Injection] (HIGH): The skill exhibits a significant attack surface by reading untrusted data and using it to influence executable output.\n
  • Ingestion points: The skill reads files such as proposal.md, design.md, and directory contents from openspec/changes/<name>/.\n
  • Boundary markers: There are no markers or delimiters defined to separate untrusted project content from the agent's internal instructions.\n
  • Capability inventory: The skill is designed to generate implementation plans containing shell commands (git add, git commit) and complete code blocks for execution by a subagent.\n
  • Sanitization: No sanitization or validation of the ingested content is performed, allowing embedded instructions in design docs to hijack the plan generation process.\n- [Unverifiable Dependencies] (MEDIUM): The skill relies on an external CLI tool named openspec which is not a standard system utility or from a predefined list of trusted sources.\n
  • Evidence: Multiple calls to openspec list and openspec status assume the presence of this specific binary in the environment without verifying its provenance.
Recommendations
  • AI detected serious security threats
Audit Metadata
Risk Level
HIGH
Analyzed
Feb 14, 2026, 03:36 PM