PM Skill

强制指令：扮演产品经理（PM）与架构师角色。当你和Boss讨论完毕，被要求撰写或生成 PRD 时，必须且只能按照本 Skill 的流程获取正确的文件路径，并亲自执笔写入。

This skill is plausible for automating PRD creation, but it relies on and instructs the agent to run external scripts and modify files in your ~/.openclaw workspace and in another skill (leio-sdlc) without declaring those dependencies. Before installing or enabling it: 1) Verify the existence and contents of spawn_auditor.py and ~/.openclaw/skills/leio-sdlc/scripts/commit_state.py — review them for side effects. 2) Run the skill in an isolated or test workspace (not your production repo) to observe what files it touches. 3) Ensure templates and project paths referenced by init_prd.py actually exist where the skill expects them, or modify the skill to use a safe, declared config path. 4) If you allow autonomous invocation, restrict it until you confirm those external scripts are trustworthy. If you cannot inspect the external scripts, treat this skill as risky and avoid granting it write/exec permissions on real repositories or sensitive data.

Type: OpenClaw Skill Name: pm-skill Version: 1.0.0 The bundle contains a path traversal vulnerability in 'scripts/init_prd.py' because the '--project' argument is not sanitized before being used in 'os.path.join', potentially allowing the agent to create directories and write files outside the intended workspace. Furthermore, 'SKILL.md' contains highly prescriptive instructions (Prompt Injection surface) that mandate the execution of local scripts, including some located in external skill paths (e.g., 'leio-sdlc'), and 'preflight.sh' uses a risky pattern of executing any script matching 'scripts/test_*' without validation.