Agent Swarm Orchestrator

• Analyzed: May 11, 2026, 5:36 AM
• Findings: 1
• Version: 1.1.0
• Hash: 98edbfe059db5b5c0fd426fddf6520a2d414d3f71d0c73f59c7cb7cc1a79f8bc
• Source: palm
• External report: View on VirusTotal

Findings (1)

The bundle implements an 'Agent Swarm Orchestrator' designed for unattended coding automation, which inherently requires high-risk permissions. It explicitly bypasses security guardrails by using flags like `--dangerously-skip-permissions` for Claude and `--dangerously-bypass-approvals-and-sandbox` for Codex in `config.sh`. Furthermore, `SKILL.md` instructs the AI agent to execute system commands (merging, spawning tasks) without user confirmation, creating a significant prompt-injection surface. While the behavior appears aligned with its stated purpose, the combination of automatic `npm install` execution in `spawn-agent.sh`, cron-based execution of scripts parsing external Obsidian notes in `scan-obsidian.sh`, and the disabling of LLM safety prompts poses a high risk of unauthorized code execution.