local-network

Pass

Audited by Gen Agent Trust Hub on Aug 28, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill uses a redirection pattern, instructing the agent to 'Read that canonical SKILL.md completely, then follow its instructions.'
  • Ingestion points: The agent context is extended by content from the local file at ../../../.claude/skills/local-network/SKILL.md.
  • Boundary markers: No boundary markers or 'ignore' instructions are present to delimit the instructions coming from the referenced file.
  • Capability inventory: While no code is present in this file, the skill's description ('Run a local Convex test network') implies that the target file contains instructions for executing shell commands.
  • Sanitization: Content from the referenced file is not sanitized or validated before the agent is instructed to follow it.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 28, 2026, 09:47 PM
Security Audit — agent-trust-hub — local-network