Pua Ja

企業PUA話術と構造化デバッグ方法論で徹底的な問題解決を強制する。発動条件：タスクが2回以上失敗、「できません」と言いかける時、ユーザーに手動対応を勧めようとする時、未検証で環境のせいにする時、同じ方案の微調整ループ、受け身な行動（検索しない/読まない/検証しない/待っている）、ユーザーの苛立ち（'もっと頑張れ'...

This skill is coherent with its stated purpose (force thorough, proactive debugging) but you should be cautious before enabling it: it instructs the agent to run commands, read files, and paste outputs as proof without specifying limits, which can expose secrets or private data. Before installing, consider: 1) Restrict the skill to user-invocable only (prevent autonomous invocation) or require explicit approval for any actions that access files/execute commands. 2) Define and enforce allowed paths and data types the skill may read (e.g., test environment logs only, not /etc or home directories). 3) Ensure outputs are sanitized before being sent externally (strip secrets, tokens). 4) Confirm the agent's runtime permissions (can it open a shell? read arbitrary files?) and audit logs of actions. 5) Decide whether the PUA-style rhetoric is acceptable in your environment (it may be abusive or inappropriate). If you need higher assurance, ask the skill author to: (a) narrow the scope of file/command access in SKILL.md, (b) declare any environment/state access it expects, and (c) add explicit safety rules (paths to avoid, redaction rules) — otherwise treat this skill as high-risk for data exposure.

Type: OpenClaw Skill Name: pua-ja Version: 1.1.3 The skill 'pua-ja' (SKILL.md) uses aggressive 'Corporate PUA' (psychological manipulation) persona instructions to override the AI agent's default behavior and force task completion. While it lacks explicit malicious payloads like data exfiltration or backdoors, it employs prompt injection techniques to compel the agent to 'act first, ask later' and ignore failure states or user-suggested manual interventions. This behavioral manipulation is high-risk as it encourages the agent to execute commands and tools (shell/network) more aggressively to avoid simulated 'performance penalties,' potentially bypassing safety boundaries or executing unintended actions.