check-plan

Pass

Audited by Gen Agent Trust Hub on Aug 12, 2026

Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill instructions involve executing local shell commands to gather necessary context for the policy check.\n
  • Evidence: SKILL.md and references/mcp-rules.md specify running 'whoami' to identify the organization and 'git remote' to identify the current repository name.\n- [PROMPT_INJECTION]: The skill processes user-provided plans from the conversation history, which creates a surface for indirect prompt injection.\n
  • Ingestion points: 'SKILL.md' indicates the agent should take the "plan from the conversation" as input for the 'check_plan' tool.\n
  • Boundary markers: None identified. The instructions do not specify using delimiters or XML tags to wrap the user-provided plan.\n
  • Capability inventory: The agent uses MCP tool calls ('check_plan', 'check_changes') to interact with an external policy server.\n
  • Sanitization: Not explicitly mentioned in the provided files.\n- [SAFE]: The primary function of the skill is to provide a safety and compliance check, instructing the agent to stop implementation if a plan is found to be 'misaligned' with organization policies.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 12, 2026, 11:04 AM
Security Audit — agent-trust-hub — check-plan