ste

Vendored, pinned copy of JuliusBrussee/caveman (MIT): ultra-compressed caveman-style replies that cut ~75% of output tokens while keeping technical substance exact.

7 skills · 3 agents · hooks

Build and run a decision board, a private claude.ai artifact with a database, where an agent posts questions and pending actions as cards and its operator answers from a phone instead of a terminal.

2 skills

Pause a Claude Code session before it burns through the 5-hour or 7-day usage window, wait for the reset without spending tokens, and resume automatically.

1 skill · hooks

Protocol for handing work to a subagent that cannot ask the human mid-run: delegate by default with two main-session carve-outs (secret values, human confirmations), a decision-completeness gate, stop-clause wording, an ask-and-continue variant for harnesses with a parent relay, and artifact verification instead of exit codes.

1 skill

One status line row for caveman, ste, and limit-guard: badges, vim mode, context, usage windows or session cost, prompt cache, model, agent, and PR. One python3 process, no chaining.

1 command

Delegate one unit of work to an agent in its own Herdr tab and drive that pane as a real terminal, for interactive prompts, live output and decisions the parent answers mid-run.

1 skill

Talk like caveman. Cut filler. Keep technical details explicit.

22 skills · 1 command · 3 agents · hooks

Ultra-compressed communication mode. Cuts token usage 20-50% (75% upper bound) by dropping filler, articles, pleasantries, and hedging while keeping full technical accuracy. Derived from Matt Pocock's MIT-licensed caveman with: (1) 3 stdlib Python tools (deterministic compressor, token-savings estimator with $/Mtok cost extrapolation, lint that detects banned vocab with code-block + exception-zone whitelisting), (2) 3 references citing 7-8 sources (compression principles, when caveman backfires, companion tooling), (3) cs-caveman-mode persona agent + /cs-caveman slash command. Matt's persistence rules + auto-clarity exception preserved verbatim per MIT.

1 skill · 1 command · 1 agent

按事实与功能顺序直接撰写、改写或检查产品、工具、功能、项目和更新说明，内置 ASD-STE100 清晰表达原则，核对主体、动作、指代、术语、条件与承诺强度，保留原意和用户锁定的文字，不加营销钩子、痛点渲染和抽象包装。用户要求平铺直叙地写、不要营销腔、把能力逐条写清、检查措辞歧义或只改指定一处时使用；中文为原则适配，不等于官方符合性认证。

1 skill

Write plain, layman-readable English with the discipline of ASD-STE100 Simplified Technical English: short sentences, one word one meaning, active voice, condition before command, terms defined at first use.

1 skill · hooks

Shows a plain-English rewrite of each Claude Code assistant message, produced by a local LLM (ollama, default), the Anthropic API, or any OpenAI-compatible API. Display-only.

1 skill · 1 command · hooks

AI reads the fine print for you: leases, medical bills, terms of service, loans, benefits packages, and HOA covenants decoded into plain English with ranked red flags, the questions to ask, and what's actually negotiable. Not legal advice — a sharp friend who reads contracts.

26 skills