Back to skill

Security audit

arkcli-helper

Security checks for vulnerabilities and agentic risk

Overview

The skill is marked suspicious because its documentation recommends running unpinned remote installer scripts, while scanners also flagged references to agent configuration and credential files; the review is incomplete because some local references could not be resolved.

Consider avoiding installation until the unpinned package and remote installer can be reviewed and verified. The scanners inspected all three listed components, but their analysis is marked partial because several local references could not be resolved; therefore the evidence is incomplete and does not establish that the skill is safe.

Vulnerability Patterns
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
  • Agent Memory PoisoningWrites attacker-controlled rules into memory that affect later sessions
  • Remote Payload Retrieval and ExecutionFetches external code whose behavior can change after review
  • Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
  • Tool MisuseTool Parameter Abuse, Chaining Abuse, Unsafe Defaults
  • Rogue AgentSelf-Modification, Session Persistence
Findings (12)

Agent Config Directory Access

High
Category
Agent Snooping
Confidence
90% confidence
Finding

Skill reads from agent configuration directories (.claude/, .codex/, .gemini/). These directories may contain API keys, personal settings, and other credentials that the skill has no legitimate need to access.

Content

Scanner excerpt · SKILL.md (reported line 187)May include surrounding context.

md
odex` / `opencode` / `openclaw` / `trae` / `deepseek-harness` / `zcode` / `workbuddy`。CUA 是 Skill 而非 MCP,还可精确安装给 `hermes` / `pi`;命令只传单个 target,不扫描或改动其他 Agent。
- 本 skill 会被 `arkcli +connect` 装进很多 agent(cursor / gemini-cli / codex …40+),但 **MCP 注入只支持上一条列出的 8 个 target**。host 是其它 agent 时:要么用户点名其一作 target,要么命令会报"请显式指定" —— **绝不静默配错对象**。
- `codex` 支持 model/provider + MCP。默认 **profile 模式**写 `~/.codex/arkcli.config.toml`,需用 `codex --profile arkcli` 启动 terminal/TUI 才生效;传 `--codex-config-scope global` 才写 `~/.codex/config.toml`,该范围可能被 Codex CLI/TUI、Codex App、IDE extension 共享读取。
- `trae`(AI IDE)是 MCP-only:不配 model/provider;**无运行态宿主检测**(不会被自动当成 host),只能显式 `arkcli helper mcp trae`。默认写用户级 `~/.trae/mcp.json`,加 `--scope project` 写项目级 `./.trae/mcp.json`(项目级需在 Trae「设置 → MCP」开启「启用项目级 MCP」开关 + 重开项目)。豆包搜索成功后另写项目根 `.trae/hooks.json`,还需在 `TRAE Settings > Hooks` 启用当前项目。
- `hermes` 支持 Plan / Platform Endpoint 的 model/provider 配置,但暂不支持 MCP 注入 → MCP 请求命中就直说"暂不支持"。
- `deepseek-harness` (DSH):默认 home scope,可直接运行 `ar

Credential Access

High
Category
Privilege Escalation
Confidence
70% confidence
Finding

Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Content

Scanner excerpt · SKILL.md (reported line 190)May include surrounding context.

md
注入 → MCP 请求命中就直说"暂不支持"。
- `deepseek-harness` (DSH):默认 home scope,可直接运行 `arkcli helper configure deepseek-harness --profile <plan>` 或 `arkcli helper mcp deepseek-harness`;若要写 profile scope,传 `--dsh-config-scope profile --dsh-profile <name>`,Helper 会确保 profile 目录存在。DSH provider/模型配置落到 `$DSH_HOME/settings.yaml` 的 `llm-pi-ai.providers.arkcli-<planType>:` 段(独立 provider,**不覆盖** `llm-deepseek:` 下的 DeepSeek 官方模型;用户可在 DSH 里切换官方 DeepSeek 与 arkcli 火山模型);默认模型写 `agent-default-model:`;API key 写入 `$DSH_HOME/.credentials.yaml` version 1 文档的 `refs.ARKCLI_<PLAN>_API_KEY`,由 DSH credentials service 按 `apiKeyEnv` 解析(不覆盖用户的 `DEEPSEEK_API_KEY`)。这是 breaking change:检测到旧版 ArkCLI 的 `llm-deepseek:` 指纹、纯 flat credentials,或 version 1 文档顶层残留的 legacy credential key 时,`configure` 不自动迁移,而是拒绝写入并要求先运行 `arkcli helper reset deepseek-harness`;`reset` 兼容清理这些旧格式,但任意自定义 `apiKeyEnv` 仍视为用户配置。settings 与 credentials 作为一个可回滚事务更新;`reset` 只删除仍保持 ArkCLI provider fingerprint 的条目、对应凭据以及仍指向它的默认模型。MCP / 原生 WebSearch 关闭落在 `$DSH_HOME/profi

Agent Config Directory Access

High
Category
Agent Snooping
Confidence
90% confidence
Finding

Skill reads from agent configuration directories (.claude/, .codex/, .gemini/). These directories may contain API keys, personal settings, and other credentials that the skill has no legitimate need to access.

Content

Scanner excerpt · references/arkcli-helper.md (reported line 39)May include surrounding context.

md
- `--ov-resource`(可选):仅整组 MCP 或 `--capability agent-memory` 使用,指定 openviking-dataplane 绑定的 OpenViking 库。显式选其它 capability 时传该 flag 会报错。
- `--scope`(可选,**仅 Trae**):`global`(默认)写用户级 `~/.trae/mcp.json`;`project` 写当前目录 `./.trae/mcp.json`。其它 agent 传 `--scope project` 会报错(一律用户级全局)。
- `--keep-native-websearch`(可选):仅整组 MCP 或 `--capability web-search` 使用。豆包搜索配置成功后保留目标 Agent 的原生 WebSearch;不传时采用推荐默认值“关闭原生 WebSearch”。
- `--codex-config-scope`(可选,**仅 Codex**):`profile`(默认)写 `~/.codex/<profile>.config.toml`;`global` 写 `~/.codex/config.toml`。
- `--codex-profile`(可选,**仅 Codex profile 模式**):Codex profile 名,默认 `arkcli`。
- `--dsh-config-scope`(可选,**仅 DeepSeek Harness**):`home`(默认)写 `$DSH_HOME/cordis.patch.yml`;`profile` 写 `$DSH_HOME/profiles/<name>/cordis.patch.yml`。
- `--dsh-profile`(可选,**仅 DSH profile 模式**):与 `--dsh-config-scope profile` 一起使用;Helper 会确保对应 profile 目录存在。

Agent Config Directory Access

High
Category
Agent Snooping
Confidence
90% confidence
Finding

Skill reads from agent configuration directories (.claude/, .codex/, .gemini/). These directories may contain API keys, personal settings, and other credentials that the skill has no legitimate need to access.

Content

Scanner excerpt · references/arkcli-helper.md (reported line 129)May include surrounding context.

md
| target | MCP 写入文件 | 重载方式 |
|--------|-------------|---------|
| claude-code | `~/.claude.json`(`mcpServers.*`);model 在另一文件 `~/.claude/settings.json`,本命令不碰 | 重启 |
| codex | 默认 profile `~/.codex/arkcli.config.toml`(`mcp_servers.*`);`--codex-config-scope global` → `~/.codex/config.toml` | profile:用 `codex --profile <name>` 启动;global:重启 Codex CLI/TUI/App/IDE |
| opencode | `~/.config/opencode/opencode.json`(`mcp.*`,merge 保留其它键) | 重启 |
| openclaw | `~/.openclaw/openclaw.json`(`mcp.servers.*` + 启用 mcporter skill) | 重启 |
| trae | 默认 `~/.trae/mcp.json`(`mcpServers.*`,与 claude 同构);`--scope project` → `./.trae/mcp.json` | 全局:去「设置 → MCP」确认 MCP 已启用 + 重启;项目级:开「启用项目级 MCP」开关 + 重开项目 |

Agent Config Directory Access

High
Category
Agent Snooping
Confidence
90% confidence
Finding

Skill reads from agent configuration directories (.claude/, .codex/, .gemini/). These directories may contain API keys, personal settings, and other credentials that the skill has no legitimate need to access.

Content

Scanner excerpt · references/arkcli-helper.md (reported line 109)May include surrounding context.

text
✓ 已为 claude-code 注入 MCP → [mcp-server-askecho-search-infinity dataPro-search openviking-dataplane openviking-controlplane]  (plan: agent-plan_cn-beijing)
✓ claude-code 原生 WebSearch 已关闭(项目级)→ /repo/.claude/settings.json
快捷恢复:arkcli helper reset claude-code(同时移除 ArkCLI 写入的模型/provider/MCP 配置)
提示:Claude Code 需重启后才会加载新注入的 MCP。

Agent Config Directory Access

High
Category
Agent Snooping
Confidence
90% confidence
Finding

Skill reads from agent configuration directories (.claude/, .codex/, .gemini/). These directories may contain API keys, personal settings, and other credentials that the skill has no legitimate need to access.

Content

Scanner excerpt · references/arkcli-helper.md (reported line 119)May include surrounding context.

text
团队版 Agent Plan 不含 OpenViking,已跳过数据面/控制面两台 MCP。
✓ 已为 claude-code 注入 MCP → [mcp-server-askecho-search-infinity dataPro-search]  (plan: agent-plan-team_cn-beijing_team)
✓ claude-code 原生 WebSearch 已关闭(项目级)→ /repo/.claude/settings.json
快捷恢复:arkcli helper reset claude-code(同时移除 ArkCLI 写入的模型/provider/MCP 配置)
提示:Claude Code 需重启后才会加载新注入的 MCP。

Agent Config Directory Access

High
Category
Agent Snooping
Confidence
90% confidence
Finding

Skill reads from agent configuration directories (.claude/, .codex/, .gemini/). These directories may contain API keys, personal settings, and other credentials that the skill has no legitimate need to access.

Content

Scanner excerpt · references/arkcli-helper.md (reported line 128)May include surrounding context.

md
| target | MCP 写入文件 | 重载方式 |
|--------|-------------|---------|
| claude-code | `~/.claude.json`(`mcpServers.*`);model 在另一文件 `~/.claude/settings.json`,本命令不碰 | 重启 |
| codex | 默认 profile `~/.codex/arkcli.config.toml`(`mcp_servers.*`);`--codex-config-scope global` → `~/.codex/config.toml` | profile:用 `codex --profile <name>` 启动;global:重启 Codex CLI/TUI/App/IDE |
| opencode | `~/.config/opencode/opencode.json`(`mcp.*`,merge 保留其它键) | 重启 |
| openclaw | `~/.openclaw/openclaw.json`(`mcp.servers.*` + 启用 mcporter skill) | 重启 |

Credential Access

High
Category
Privilege Escalation
Confidence
70% confidence
Finding

Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Content

Scanner excerpt · references/arkcli-helper.md (reported line 197)May include surrounding context.

md
| reset 未恢复原生搜索 | 用户原本就已关闭,或配置后同一字段被外部修改 | ArkCLI 不接管原有禁用;冲突时保留用户当前值并输出警告 |
| DSH 提示检测到旧版配置或 flat credentials | 旧版 ArkCLI 写入格式与当前 DSH provider / credentials 契约不兼容;flat key 可能单独存在,也可能残留在 version 1 文档顶层 | 先运行 `arkcli helper reset deepseek-harness`,再重新执行 `configure`;不要手工删除 `apiKeyEnv` |

DSH model/provider 写入还遵循以下 ownership 规则:`settings.yaml` 使用独立的 `llm-pi-ai.providers.arkcli-<planType>`,凭据使用 `.credentials.yaml` version 1 文档的 `refs.ARKCLI_<PLAN>_API_KEY`,让 DSH credentials service 能按 provider 的 `apiKeyEnv` 解析。两个文件按一个可回滚事务更新。检测到纯 flat credentials、version 1 文档顶层残留的 legacy credential key,或 `llm-deepseek` 同时具有 `apiKeyEnv: DEEPSEEK_API_KEY` 与精确 Ark 数据面 HTTPS URL 的旧 ArkCLI 指纹时,`configure` fail-fast,不做隐式迁移,并明确要求先运行 `arkcli helper reset deepseek-harness`;该 reset 兼容读取旧格式并清理旧 ArkCLI 内容。不要把自定义 `apiKeyEnv` 当成 ArkCLI ownership。`helper reset deepseek-harness` 仅移除 fingerprint 仍完整匹配的 provider、对应凭据以及仍指向该 provider 的 `agent-default-model`。
| ZCode 调用 `glm-5.2` 等模型时报 `InvalidParameter` / token 上限 400 | 旧配置把 A

Self-Modification

High
Category
Rogue Agent
Confidence
90% confidence
Finding

Skill modifies its own code, configuration, or behavior at runtime. Self-modification enables an agent to escalate privileges, disable safety constraints, or install persistent backdoors.

Content

Scanner excerpt · references/arkcli-helper.md (reported line 277)May include surrounding context.

Agent 进化当前官方适配 Claude Code / OpenClaw / Trae;安装命令为:

bash
curl -fsSL "https://ark-self-evolve.tos-cn-beijing.volces.com/evolve_skill/latest/install.sh" | bash

CLI 只展示该命令,不自动执行未固定版本的远程脚本。安装完成后返回矩阵选择“刷新状态”。

External Script Fetching

High
Category
Supply Chain
Confidence
98% confidence
Finding

The recommended manual installation command downloads a mutable latest/install.sh and immediately executes it without checking a signature or pinned digest. Not executing it automatically reduces exposure, but users following the recommendation could run attacker-controlled code if the hosting account or artifact is compromised.

Content

Scanner excerpt · references/arkcli-helper.md (reported line 277)May include surrounding context.

Agent 进化当前官方适配 Claude Code / OpenClaw / Trae;安装命令为:

bash
curl -fsSL "https://ark-self-evolve.tos-cn-beijing.volces.com/evolve_skill/latest/install.sh" | bash

CLI 只展示该命令,不自动执行未固定版本的远程脚本。安装完成后返回矩阵选择“刷新状态”。

Chaining Abuse

High
Category
Tool Misuse
Confidence
98% confidence
Finding

Piping curl output directly into bash turns fetched content into executable instructions without an intervening verification step. This is the same underlying installer weakness as SC2, not evidence of malicious chaining, and manual execution still exposes the user's files and credentials to any compromised payload.

Content

Scanner excerpt · references/arkcli-helper.md (reported line 277)May include surrounding context.

Agent 进化当前官方适配 Claude Code / OpenClaw / Trae;安装命令为:

bash
curl -fsSL "https://ark-self-evolve.tos-cn-beijing.volces.com/evolve_skill/latest/install.sh" | bash

CLI 只展示该命令,不自动执行未固定版本的远程脚本。安装完成后返回矩阵选择“刷新状态”。

MCP server referenced without pinned version: 'npx -y @byted-supabase/cli'.

Medium
Category
MCP Rug Pull
Confidence
98% confidence
Finding

The documented installation executes @byted-supabase/cli@latest through npx without a fixed version, making the executed code dependent on mutable registry content. Although this is a CLI rather than an MCP server, a compromised release could execute with the user's permissions in a workflow that subsequently handles STS and refresh-token credentials.

Content

No source excerpt is available for this finding.