Skill flagged — suspicious patterns detected

ClawHub Security flagged this skill as suspicious. Review the scan results before using.

Kimi Code CLI

v1.0.0

使用 Kimi Code CLI 执行复杂代码任务。当用户需要: 1. 大型代码生成(完整功能模块、多文件项目) 2. 复杂代码重构(legacy 代码现代化、架构调整) 3. 深度 Bug 诊断(多文件关联错误、难以定位的问题) 4. 技术调研实现(新技术栈完整集成、PoC 开发) 5. 长时间运行的开发任务(...

1· 590·0 current·1 all-time
byFlynn Guo@walkman1w
MIT-0
Download zip
LicenseMIT-0 · Free to use, modify, and redistribute. No attribution required.
Security Scan
VirusTotalVirusTotal
Suspicious
View report →
OpenClawOpenClaw
Suspicious
medium confidence
!
Purpose & Capability
The name/description (run and orchestrate Kimi CLI to generate/modify code) aligns with the content at a high level, but the package claims many artifacts (cron jobs, payload.acp, scripts/acp_check.py, swarm_consumer.py, summary.json, etc.) that are not present. The SKILL.md recommends using 'kimi --print -p ...' as the canonical one-shot form, yet the included runner builds commands using "kimi '{task}'" (no --print -p). That mismatch means the tool may enter interactive mode contrary to the documentation. Also, SKILL.md expects users to have 'kimi' installed and logged in, but the skill does not declare or request these credentials.
!
Instruction Scope
Runtime instructions direct the agent to run shell exec commands via OpenClaw 'exec' (PTY mode, background sessions, process actions). The bundled script generates shell commands by interpolating user-provided tasks directly into a quoted shell command, which can produce unexpected interactive sessions or enable shell injection if task text contains quotes or shell metacharacters. The script also offers an optional 'wake' command that runs 'openclaw gateway wake ...' (external notification); that causes network/activity outside the local workdir and is not described in requires.env. The SKILL.md's version notes mention cron jobs and periodic checks, but no install or setup steps create them — the instructions therefore claim behaviors that the repository does not implement.
Install Mechanism
There is no install spec (instruction-only + one small helper script). That minimizes installation risk since nothing is downloaded automatically. However, SKILL.md expects 'pip install kimi-cli' and 'kimi /login' to have been done by the user; those are external prerequisites not enforced or documented as environment requirements here.
!
Credentials
The skill declares no required environment variables or primary credential, yet operationally needs external tooling and authentication: (1) the Kimi CLI requires the user to install and log in ('kimi /login'); (2) the optional wake command calls 'openclaw gateway wake' which likely requires gateway authorization. Those credentials are not declared or scoped. This mismatch (no declared secrets but implicit external auth) is disproportionate and hides what secrets the skill will rely on or could expose.
Persistence & Privilege
always is false and the skill does not request elevated privileges or modify other skills. However, the SKILL.md and changelog claim periodic cron jobs and hourly checks; the repository contains no installation steps to create such persistence. That discrepancy is suspicious but not an immediate privilege escalation by itself.
What to consider before installing
High-level checklist before using/installing: - Ask the author/source for provenance: where does 'kimi' come from, and what credentials (if any) does it require? The package claims files (acp_check.py, cron setup, summary outputs) that are not included — ask why. - Do not run generated commands with real secrets or on sensitive hosts. The helper script interpolates your task string directly into a shell command (potential shell-injection risk). Sanitize or constrain inputs and run first in an isolated container or throwaway VM. - Expect to manually install and log into the external Kimi CLI (pip install kimi-cli; kimi /login). The skill does not declare or manage those credentials — think about where those tokens/credentials live and whether they should be provided to the agent. - Be cautious with the optional 'wake' feature: it runs 'openclaw gateway wake' which triggers network activity outside the workdir; verify what that does and whether it requires additional auth. - If you plan to use background/interactive mode, follow SKILL.md advice to create an isolated workdir beforehand. Test quick one-shot commands with a safe, simple task to see whether they spawn an interactive session unexpectedly (the runner currently uses 'kimi '{task}'' which may be interactive). If you want a safer go/no-go: request the missing files and a clear statement of required credentials (and how they are stored), or run the skill only in an isolated environment after those clarifications.

Like a lobster shell, security has layers — review code before you run it.

coding kimi cli agent developmentvk97489jz5h39jm3s7g046ydypd81endxlatestvk97489jz5h39jm3s7g046ydypd81endx

License

MIT-0
Free to use, modify, and redistribute. No attribution required.

SKILL.md

Kimi CLI Skill

版本: 1.2.4 | 最后更新: 2026-02-04

使用 Kimi Code CLI 作为“编码执行器”,配合会话模型做规划/验收,完成复杂代码任务。

架构与职责分工(重要)

目标:不是让会话模型(例如 GLM)直接写复杂代码,而是让它指挥,由 Kimi Code CLI 真正产出/修改代码文件

🧠 会话模型(规划/拆解/验收)
  ↓ 调用
📋 kimi-cli skill(exec + pty + workdir + process 控制)
  ↓ 运行
💻 Kimi Code CLI(生成/修改文件、输出步骤)
  • 会话模型:需求澄清、任务拆解、验收标准、结果核查、复盘总结
  • Kimi CLI:按指令在指定目录里创建/修改文件、生成脚本/项目、给出执行步骤

两种工作模式

1. 快速模式 (Quick Mode)

适用场景:明确的单次任务,执行完即返回

推荐执行方式(one-shot)

bash pty:true workdir:~/project timeout:300 command:"kimi --print -p '你的任务描述'"

参数说明

  • pty:true - 必需! 分配伪终端,让 CLI 输出/交互稳定
  • workdir - 工作目录(强烈建议使用独立目录)
  • timeout - 超时时间(秒)
  • command - 要执行的 Kimi 命令

示例

# 创建一个完整项目
bash pty:true workdir:~/projects timeout:600 command:"kimi --print -p '创建一个 React + TypeScript 的待办事项应用'"

# 重构现有代码
bash pty:true workdir:~/my-project timeout:300 command:"kimi --print -p '重构 src/utils.py 使用现代 Python 最佳实践'"

注意:不要用 kimi '...' 作为 one-shot 标准写法(容易进入交互/输出不一致)。


2. 交互模式 (Interactive Mode)

适用场景:复杂任务需要多轮交互,或 Kimi 可能会提问确认

启动方式

bash pty:true workdir:~/project background:true command:"kimi"

后续控制(使用 process 工具):

操作命令
查看输出process action:log sessionId:xxx
发送输入process action:submit sessionId:xxx data:"你的回答"
原始输入process action:write sessionId:xxx data:"内容"
检查状态process action:poll sessionId:xxx
结束会话process action:kill sessionId:xxx

完整示例

# 1. 启动后台会话
bash pty:true workdir:~/project background:true command:"kimi"
# 返回: sessionId: abc123

# 2. 查看初始输出
process action:log sessionId:abc123

# 3. Kimi 提问时,发送回答
process action:submit sessionId:abc123 data:"使用 TypeScript"

# 4. 持续监控
process action:log sessionId:abc123 limit:50

# 5. 完成或出错时结束
process action:kill sessionId:abc123

Prompt 最佳实践

基于测试任务总结的最佳实践:

1. 明确指定命令版本

许多系统同时安装了 pythonpython3,但 python 可能不存在:

# ✅ 明确指定 python3
kimi --print -p "使用 python3 运行..."

# ❌ 不指定可能失败
kimi --print -p "使用 python 运行..."  # 可能找不到命令

2. 要求包含测试 / 验证步骤

kimi --print -p "创建一个模块,要求:1) 功能实现 2) 包含单元测试文件 3) 给出如何运行 pytest 的命令"

3. 分步骤描述需求

kimi --print -p "创建一个脚本,要求:1) 支持参数 X 2) 验证输入 3) 错误处理 4) 文档"

最佳实践(执行层)

1. 总是使用 PTY

Kimi CLI 是交互式应用,必须使用 pty:true

# ✅ 正确
bash pty:true command:"kimi --print -p '...'"

# ❌ 错误 - 可能挂起或输出异常
bash command:"kimi --print -p '...'"

2. 使用独立工作目录(并确保目录已存在)

避免 Kimi 访问不相关文件:

bash pty:true workdir:/tmp/new-project command:"kimi --print -p '创建新项目'"

注意(OpenClaw exec 的 workdir 约束)

  • execworkdir 必须在调用前就存在,否则 OpenClaw 会回退到默认目录(并打印类似 “workdir ... is unavailable”)。
  • 推荐做法:先单独创建目录,再执行 Kimi。

示例:

# 1) 先创建目录(普通 exec 即可)
mkdir -p /tmp/kimi-eval/t01

# 2) 再用该 workdir 调用 Kimi(PTY 必开)
bash pty:true workdir:/tmp/kimi-eval/t01 timeout:600 command:"kimi --print -p '...任务...'"

3. 合理设置超时

# 小任务: 5 分钟
bash pty:true timeout:300 command:"kimi --print -p '...'"

# 大任务: 10 分钟或更长
bash pty:true timeout:600 command:"kimi --print -p '...'"

4. 监控后台任务

process action:log sessionId:xxx limit:30

5. 处理 Kimi 的提问

process action:submit sessionId:xxx data:"y"

错误处理:429 Rate Limit(必须掌握)

当 Kimi 配额耗尽时,可能出现:

  • 429 ... rate_limit_error
  • 表现为:子任务“无输出”/多次重试仍失败

SOP:

  1. 识别:日志/错误信息包含 429rate_limit_error
  2. 停止重试:短时间重复调用只会继续失败
  3. 等待:通常按“下个计费/刷新周期”恢复(建议 60 分钟后再试)
  4. 记录:把受影响任务标注为 blocked_by_rate_limit,避免误判为失败
  5. 恢复后补跑:配额恢复后优先补跑被阻塞任务

可选降级:如果必须立刻继续推进,可由会话模型临时代写/生成草案,但要标注为“非 Kimi 产物”,配额恢复后再用 Kimi 复跑对比。


依赖要求

  • Kimi Code CLI 已安装:pip install kimi-cli
  • 已登录:kimi /login

版本历史

v1.2.4 (2026-02-04)

改进点:

  • ✅ 完成 A03 (ACP 合规输出):Kimi 生成 payload.acp(acpVersion=2.0, signature, data),scripts/acp_check.py 验证通过(VALID)
  • ✅ 设置每小时自动检查任务(cron job),完成上一题后自动启动下一题
  • ✅ 新增 5 道探索性题目(E01-E05):自我监控、渐进式复杂度、多语言混合、失败恢复、网络降级

v1.2.3 (2026-02-04)

改进点:

  • ✅ 完成 R01 (Ralph print) 验证:Kimi 成功生成 summary.json + summary.txt,数据一致(metric=42, status=SUCCESS),JSON 通过 python3 -m json.tool 校验
  • ✅ 确认 Ralph-style 输出格式可被后续 agent 解析
  • ✅ 完成 S02 (Agent Swarm):验证 exec → config.json → swarm_consumer.py → swarm-report.md 链路

v1.2.2 (2026-02-04)

改进点:

  • ✅ 增补“mini_kv + pytest”完整迭代案例(T02),并记录:虚拟环境/依赖安装限制 + pip3 + --break-system-packages 解决办法
  • ✅ 强调 pip install 可能因 PEP 668/externally-managed 限制失败,需使用 pip3 或系统包 + --break-system-packages 作为备选

v1.2.1 (2026-02-04)

改进点:

  • ✅ 补充 exec.workdir 需要“目录预先存在”的注意事项与推荐流程(避免 silently fallback)

v1.2.0 (2026-02-04)

改进点:

  • ✅ 统一 Quick Mode 的 one-shot 命令为 kimi --print -p "..."(移除 kimi '...' 误导写法)
  • ✅ 增加“会话模型 + kimi-cli + Kimi CLI”的架构与职责分工说明
  • ✅ 增加 429 rate limit 的处理 SOP(识别/等待/记录/补跑)
  • ✅ 更新测试计划进度(L1 已完成)

v1.1.0 (2026-02-04)

改进点:

  • ✅ 修正命令格式为 kimi --print -p "prompt"
  • ✅ 添加明确指定 python3 的最佳实践(避免环境问题)
  • ✅ 添加 Prompt 最佳实践章节
  • ✅ 创建 50 任务测试套件(L1-L5)

v1.0.0 (2026-02-04)

初始版本:

  • 基础 PTY 模式支持
  • Quick Mode 和 Interactive Mode
  • 文档结构建立

测试计划(50 任务)

目标:50 道测试任务,迭代优化 Skill

阶段任务数状态
L1 基础10✅ 完成 (10/10)
L2 中等15待开始
L3 复杂15待开始
L4 交互5待开始
L5 调试5待开始

当前进度:20% (10/50)

Files

3 total
Select a file
Select a file to preview.

Comments

Loading comments…