Pressure skill
Skill Someone-hates-Monday/pressure-skill/.agents/skills/pressure-skill
Cold pressure → warm, sendable replies (ZH/EN). Layered workplace comms—deflect, nudge, pin vague specs—like a super mouthpiece that won't get you fired. Multi-platform Agent Skill (Cursor / Claude Code / OpenClaw) + FastAPI; offline templates. ZH: 拒绝被压力、拒绝职场PUA、反催办、不接锅、软拒绝、立边界。 EN: pushback, boundaries, anti-gaslighting, say-no.From its SKILL.md
npx -y skills add Someone-hates-Monday/pressure-skill --skill pressure-skillAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
2 things to look at
- 3 stars3 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
- runs commandsInstructs the agent to run 1 command, including `counterparty_cli.py feedback`.
SKILL.md
4.8 KB, ~1.6k tokens by cl100k_base, as published. Nobody here has run it
Language / 语言
Supports English and Chinese (same pattern as many open skills): infer from the user’s first substantive message and stay in that language for questions and drafts unless they switch explicitly.
支持中文与英文:根据用户第一条实质性消息判定语言,并全程使用同一语言;用户若明确切换语言则跟随。
pressure.skill(多平台 Agent Skill)
轻松一句:冰冷压力 → 温暖回复;像一套立体攻防替你守住边界,再当你的超级嘴替——把「再催我就鼠掉」翻译成对方听得进、你发得出手的版本。
Agent 执行协议(最高优先级)
- 定位技能根目录
<skill-root>:即 本SKILL.md所在目录(无论装在仓库内.cursor/skills/pressure-skill/、.agents/skills/pressure-skill/,还是用户目录下的~/.cursor/skills/pressure-skill、~/.claude/skills/pressure-skill、~/.openclaw/workspace/skills/pressure-skill等)。 - 用户挂上本 Skill 后的第一条行动:用当前产品支持的读文件能力读取
<skill-root>/references/wizard-full-flow.md- Cursor:工具名一般为 Read。
- Claude Code / OpenClaw:使用文档中与 Read 等价的文件读取工具(名称以各版本为准)。
可选:先读<skill-root>/references/platform-wizard-notes.md,了解多平台路径与 Python cwd 约定。
- 若读取失败(工作区未含该路径、或沙箱无法读盘):请用户 将 pressure.skill 仓库克隆根打开为工作区、提供
wizard-full-flow.md的磁盘路径,或 将wizard-full-flow.md全文粘贴进对话;在取得流程前 不要编造向导步骤。 - 读完主向导后:严格按
wizard-full-flow.md章节顺序与用户多轮对话;除「快速通道」外,禁止在 画像小结(§4)经用户确认前 采集场景 A/B/C/D,禁止在 用户声明目的(§5)前 输出最终可发送话术定稿。 - 长期对象(复诊):用户要继续跟踪同一位对方、追加材料或反馈时,再 Read
<skill-root>/references/counterparty-long-term.md;经用户同意后用仓库根scripts/counterparty_cli.py读写本地counterparties/(默认不进 Git)。复诊时仍须每轮采集 §5 目的 与 §6~7 新场景。
5b. 实地反馈(发出去之后):用户带回真实对话与对方反应、或澄清了模糊话的真义时,再 Read<skill-root>/references/counterparty-feedback-loop.md,对照上轮预测写偏差与learned_rules,执行counterparty_cli.py feedback校准画像,使下次意图推测与话术更准。 - 收尾顺序:遵循向导 §10——先问用户此刻最需要哪类产出并置顶作答,再写意图推测、多假设意图、话术及对方可能反应;科研/模糊对齐场景可按向导加载
domain-artefact-cheatsheet.md;深证据(readiness.portrait_depth=deep)时按向导加载evidence-based-reply-memo.md。 - 语言:遵循
wizard-full-flow.md中的 Language / 语言 规则(中英一致)。
本文件其余内容(快速备忘)
- 隐私:不默认写入仓库;提醒用户脱敏。
- 双路径:对话内可直接给建议(由当前产品绑定的模型执行);是否在仓库根跑
bundle_local.py由 Agent 按向导 §8 自行判断(不向用户问工具名);若跑,在用户话术里只用自然语言交代「做了信息结构化检查」。若技能只装在用户目录,向导仍在该<skill-root>/references/下;Python 脚本(bundle_local.py、counterparty_cli.py等)必须在 pressure.skill 仓库克隆根(含scripts/)执行。详见platform-wizard-notes.md与向导 §8~§9、长期档案counterparty-long-term.md。 - readiness:脚本/API 会输出信息是否够用;向导 §10 规定收尾顺序(先问产出优先级 → 再意图推测/话术/风险/追问)。
- 工作目录:运行脚本时 cwd 须为含
scripts/的仓库根,不是<skill-root>。 - 多平台安装 / 排障:见 docs/skill-adapters.md。
- HTTP:
uvicorn app:app,接口见仓库README.md。
What ships with it: 6 files
33.5 KB alongside SKILL.md