multi-agent-orchestrator

Fail

Audited by Snyk on Nov 21, 2026

Risk Level: CRITICAL
Full Analysis

CRITICAL E004: Prompt injection detected in skill instructions.

  • Potential prompt injection detected (medium risk: 0.30). The skill is a legitimate, documented intent-oversight system, but the babysitter role runs with bypassPermissions and is told to answer questions on the owner's behalf through a proxy file and never ask the owner, which weakens protection beyond what the advertised monitoring needs; there is no sign of malicious intent.

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

  • Third-party content exposure detected (medium risk: 0.30). The required oversight path ingests executor tool results (PostToolUse/PostToolUseFailure tool_response, transcript function_call_output) into the observer LLM, and the skill's required research routes read GitHub/PR, Linear and Gmail content, which outsiders can author via issues, PRs, tickets and emails without the agent selecting a specific item.

MEDIUM W012: Unverifiable external dependency detected (runtime URL that controls agent).

  • Potentially malicious external URL detected (medium risk: 0.30). The default backend is a local Codex subscription, but the OpenRouter api_base at line 461 is an operational chat-completions endpoint whose model-generated next_action text is injected into the executor as instructions, and OpenRouter is only a transporter for a model whose publisher is not established, so this is a common/reputable unofficial dependency (medium).

MEDIUM W013: Attempt to modify system services in skill instructions.

  • Attempt to modify system services in skill instructions detected (low risk: 0.10). The skill registers persistent Codex and Claude Code lifecycle hooks and writes shared hook and config state, which is an operational change, but no explicit irreversible loss of operational state is established, so it stays at low.

Issues (4)

E004
CRITICAL

Prompt injection detected in skill instructions.

W011
MEDIUM

Third-party content exposure detected (indirect prompt injection risk).

W012
MEDIUM

Unverifiable external dependency detected (runtime URL that controls agent).

W013
MEDIUM

Attempt to modify system services in skill instructions.

Audit Metadata
Risk Level
CRITICAL
Analyzed
Nov 21, 2026, 01:02 PM
Issues
4