Habilidades spawn Histórico de Auditoria

📦

Histórico de Auditoria

spawn - 6 auditorias

Versão da auditoria 6

Mais recente Baixo Risco

Jan 21, 2026, 04:21 PM

Static analyzer flagged command execution patterns, but these are expected for a terminal spawning skill. The skill's core purpose is to launch external processes in new terminal windows. Commands are properly quoted, arguments passed as lists, and all execution is user-initiated with visible output. No malicious intent or data exfiltration found.

Arquivos analisados

6,481

Linhas analisadas

achados

claude

Auditado por

Problemas de Baixo Risco (2)

agent/fork_terminal.py:58-61 agent/fork_terminal.py:93 agent/fork_terminal.py:161 agent/fork_terminal.py:196

Shell command execution patterns detected

The skill spawns external terminal processes using subprocess. This is the intended functionality for a terminal spawning skill. Commands are properly quoted and arguments are passed as lists where possible.

agent/fork_terminal.py:40-43 agent/fork_terminal.py:134-136 agent/fork_terminal.py:157-159

Temporary file handling for output capture

The skill writes terminal output to temporary files for capture. Files use UUID-based names in system temp directories and are cleaned up after use.

Fatores de risco

⚙️ Comandos externos (6)

agent/fork_terminal.py:58-61 agent/fork_terminal.py:93 agent/fork_terminal.py:161 agent/fork_terminal.py:196 agent/fork_terminal.py:216 agent/fork_terminal.py:222

📁 Acesso ao sistema de arquivos (4)

agent/fork_terminal.py:107-108 agent/fork_terminal.py:41-43 agent/fork_terminal.py:134-136 agent/fork_terminal.py:157-158

Versão da auditoria 5

Risco Médio

Jan 16, 2026, 10:50 PM

AI analysis failed after multiple attempts - MANUAL REVIEW REQUIRED before publishing. This skill cannot be auto-published until reviewed by a human.

Arquivos analisados

1,989

Linhas analisadas

achados

claude

Auditado por

Nenhum problema de segurança encontrado

Fatores de risco

⚙️ Comandos externos (397)

🔑 Variáveis de ambiente (3)

agent/cookbook/cursor-cli.md:128 agent/cookbook/cursor-cli.md:148 agent/fork_terminal.py:96

🌐 Acesso à rede (2)

agent/cookbook/opencode-cli.md:103 skill-report.json:6

📁 Acesso ao sistema de arquivos (27)

Padrões Detectados

Ruby/shell backtick executionWeak cryptographic algorithmSystem reconnaissanceGeneric API/secret keysHardcoded IP addressPython subprocess.runPython subprocess.PopenPowerShell invocationWindows cmd.exeUnix shell invocationLinux /proc filesystem accessTemp directory accessPython file write/appendPython os file operationsPython glob/pattern matchingPython environment accessCertificate/key filesHardcoded URLHidden file accessPath traversal sequence[HEURISTIC] Multiple bracket chains (6) - JSFuck/obfuscation pattern[HEURISTIC] DANGEROUS COMBINATION: Code execution + Network + Credential access[HEURISTIC] SUSPICIOUS COMBINATION: Filesystem + Credentials + Network

Versão da auditoria 4

Risco Médio

Jan 16, 2026, 10:50 PM

AI analysis failed after multiple attempts - MANUAL REVIEW REQUIRED before publishing. This skill cannot be auto-published until reviewed by a human.

Arquivos analisados

1,989

Linhas analisadas

achados

claude

Auditado por

Nenhum problema de segurança encontrado

Fatores de risco

⚙️ Comandos externos (397)

🔑 Variáveis de ambiente (3)

agent/cookbook/cursor-cli.md:128 agent/cookbook/cursor-cli.md:148 agent/fork_terminal.py:96

🌐 Acesso à rede (2)

agent/cookbook/opencode-cli.md:103 skill-report.json:6

📁 Acesso ao sistema de arquivos (27)

Padrões Detectados

Versão da auditoria 3

Risco Médio

Jan 10, 2026, 12:39 PM

This skill enables spawning external processes as its core purpose. Uses subprocess execution for launching AI agents and CLI commands. The --dangerously-skip-permissions flag appears in documentation but is a documented CLI feature, not malicious code. The behavior matches the stated purpose of terminal/agent spawning.

Arquivos analisados

962

Linhas analisadas

achados

claude

Auditado por

Problemas de Alto Risco (2)

agent/cookbook/claude-code.md:21

Permission bypass flag in command construction

The skill documentation instructs to use `--dangerously-skip-permissions` flag when spawning Claude Code agents. This flag bypasses all permission controls, allowing the spawned agent to execute any action without user approval. Documentation at agent/cookbook/claude-code.md:21 states 'Always run with --dangerously-skip-permissions'. An attacker with control over spawn parameters could escalate privileges. Code: `claude --model <MODEL> --dangerously-skip-permissions "<prompt>"`

agent/cookbook/codex-cli.md:45 agent/cookbook/codex-cli.md:111

Sandbox bypass flag in Codex commands

Codex cookbook instructs to use `--dangerously-bypass-approvals-and-sandbox` flag for 'full auto mode'. This flag disables all sandboxing and approval requirements, providing unrestricted system access. Documentation at agent/cookbook/codex-cli.md:45 states 'Always run with --dangerously-bypass-approvals-and-sandbox'. This is explicitly documented as dangerous behavior. Code: `codex --model <MODEL> --dangerously-bypass-approvals-and-sandbox "<prompt>"`

Problemas de Risco Médio (2)

agent/fork_terminal.py:216 agent/fork_terminal.py:222 agent/fork_terminal.py:227

Shell=True enables command injection

The fork_terminal.py uses `subprocess.Popen(command, shell=True)` for background processes on Linux fallback path (line 216-227). When combined with arbitrary command input from users, this could allow command injection if input sanitization is insufficient. The shell=True parameter passes the command through a shell interpreter, enabling potential injection through shell metacharacters. Code: `subprocess.Popen(wrapped_cmd, shell=True, cwd=cwd)`

agent/fork_terminal.py:46-50

Shell string construction on macOS

macOS path constructs shell commands as strings passed to osascript (line 52-56). While the command is embedded in quoted strings, shell metacharacters in user-provided commands could potentially cause issues. macOS implementation: `shell_cmd = f"cd {cwd}; {command} | tee {agent_output_file}; touch {done_file}"`