Bullybuddy

• Analyzed: May 14, 2026, 3:51 AM
• Findings: 1
• Version: 0.7.1
• Hash: 8121fcc44c8af509c926be4f8be32681b9e57ccdfdf2898a064598d85dcc86c5
• Source: palm
• External report: View on VirusTotal

Findings (1)

The `scripts/bullybuddy.sh` script is vulnerable to JSON injection in its `spawn` and `send` commands, as user-supplied arguments (`$cwd`, `$task`, `$text`) are directly interpolated into JSON payloads without proper escaping. This could allow an attacker to inject arbitrary JSON fields into API requests. Furthermore, the skill exposes a high-risk capability via the `/bullybuddy send` command, which allows arbitrary command execution within managed Claude Code sessions, a feature explicitly highlighted in `SKILL.md` as granting 'full control'. The `command-arg-mode: raw` in `SKILL.md` exacerbates this, making the skill highly susceptible to prompt injection against the OpenClaw agent, which could then be coerced into exploiting these vulnerabilities or capabilities.