auto-plan

Pass

Audited by Gen Agent Trust Hub on Aug 11, 2026

Risk Level: SAFE
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from specifications (via the spec argument or .ulpi/spec/* paths) to generate build plans. This attack surface is well-defended. \n
  • Ingestion points: Specification files provided as input.\n
  • Boundary markers: Content is processed into a structured JSON schema.\n
  • Capability inventory: Uses Bash, Write, Agent, and Workflow tools to generate plans for subsequent automated building.\n
  • Sanitization: The scripts/validate-plan.mjs script performs strict regex validation (^[A-Za-z0-9][A-Za-z0-9_-]*$) on task identifiers to prevent shell metacharacter injection and directory traversal when the plan is executed by downstream build agents. It also utilizes an adversarial review loop to verify the logic and safety of the generated task graph.\n- [COMMAND_EXECUTION]: The skill executes a local validation script (scripts/validate-plan.mjs) using the Bash tool. This script is part of the skill's own package and serves as a security gate to enforce structural and safety invariants on the generated output.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 11, 2026, 06:20 PM
Security Audit — agent-trust-hub — auto-plan