visual-plan

Pass

Audited by Gen Agent Trust Hub on Sep 11, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill relies on the '@agent-native/core' package, which is fetched and executed via npx or bunx to provide the planning and visualization engine.
  • [COMMAND_EXECUTION]: The skill instructions include executing shell commands for tool setup, session reconnection, and local plan verification using the '@agent-native/core' CLI.
  • [INDIRECT_PROMPT_INJECTION]: The skill features an indirect injection surface by ingesting external human feedback via the 'get-plan-feedback' tool.
  • Ingestion points: Human feedback threads and review event payloads (references/agent-native-plan.md).
  • Boundary markers: Use of 'resolutionTarget' and 'anchorDetails' to distinguish agent-targeted from human-targeted feedback.
  • Capability inventory: Tool-based file/data writing via 'update-visual-plan' and shell command execution.
  • Sanitization: Documentation explicitly mentions that output is 're-sanitized' during content patching.
  • [DATA_EXFILTRATION]: Plan content is shared with 'https://plan.agent-native.com' to enable hosted collaboration and commenting. This is a documented core feature, and the skill provides a 'Local-Files Privacy Mode' to prevent external data transmission when required.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 11, 2026, 02:01 AM
Security Audit — agent-trust-hub — visual-plan