from talk-normal1,636
A tested, bilingual system-prompt installer that reduces verbose, corporate-sounding LLM output while preserving information.
A curated system prompt and installer that injects concise-style rules into an OpenClaw workspace's AGENTS.md. Once installed, the rules (from prompt.md) run as an always-on behavior modifier so the agent avoids filler, hedging, and corporate tics and produces shorter, clearer replies. The project includes measured benchmark data showing substantial character reduction on modern LLMs.
Use this when you want to make an agent consistently shorter and less verbose across all responses (workspace-wide), especially for teams who need direct, information-dense outputs or bilingual (English/Chinese) handling. It's a one-time install that persists until uninstalled.
Works with any LLM or agent system that respects system prompts/custom instructions (GPT-5.4, GPT-4o-mini, Claude 4.6, Gemini, etc.). Integration is via AGENTS.md injection, so it is compatible with OpenClaw-style agents and similar agent frameworks.
A system prompt installer for Hermes agents that reduces LLM verbosity. No bundled scripts were available for testing — the skill relies on an external install.sh fetched from GitHub. SKILL.md is well-documented with clear procedures, verification steps, and pitfalls. The skill transparently modifies persistent config files (AGENTS.md), which its own security scanner flags as dangerous but is the intended behavior.
No scripts bundled for execution testing. install.sh is referenced but not included — a security review of that script would require fetching from the GitHub repo. The skill's purpose is legitimate and transparent. Measured results (56-73% reduction) are cited with benchmarks.
