optimize-agent-prompt

Pass

Audited by Gen Agent Trust Hub on Feb 23, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill recommends installing the '@sanity/agent-context-explorer' package from the npm registry. This is a first-party tool from the verified vendor 'sanity-io'.
  • [COMMAND_EXECUTION]: The skill documentation includes a command to run the 'agent-context-explorer' CLI tool. This command follows security best practices by using environment variables ($SANITY_API_READ_TOKEN, $ANTHROPIC_API_KEY) to handle sensitive credentials rather than hardcoding them.
  • [PROMPT_INJECTION]: The skill defines an architecture for agents to process external data, creating a potential surface for indirect prompt injection.
  • Ingestion points: Untrusted data from Sanity datasets enters the context via an MCP server connection.
  • Boundary markers: The skill explicitly recommends a 'two-surface architecture' to separate dataset knowledge from the behavioral system prompt.
  • Capability inventory: No active scripts or executable code are contained within the skill file itself; it is primarily instructional.
  • Sanitization: The provided prompt templates include guardrails that instruct the agent to avoid guessing and to redirect users when information is unavailable.
Audit Metadata
Risk Level
SAFE
Analyzed
Feb 23, 2026, 09:17 PM