// the take
A massive awesome-list of LLM and agent prompts: a 2,790-entry bilingual (EN/ZH) everyday-prompt library plus hundreds of individual system prompts for specific agent harnesses (Claude Code, Codex, Gemini CLI, Cline, etc.), prompt-engineering techniques, and papers. It's aimed at developers and prompt engineers who want copy-paste starting points rather than people building original prompt-engineering tooling.
The everyday-prompt library is generated from a tracked CSV via a build script (scripts/build_library.py), so the category tables and counts are reproducible rather than hand-maintained. Coverage of the agent-harness ecosystem is genuinely current — individual prompts are tied to specific named projects and dates (e.g., specific CLI tools, SKILL.md formats, A2A protocol version), so it functions as a decent index of what's happening across coding-agent harnesses. A CI workflow (check-internal-links.yml) at least verifies links don't rot. Prompts are plain text/markdown files with direct raw links, so they're trivially copy-pasteable without cloning.
This is a link index, not a tool — there's no way to evaluate whether any given prompt actually works well, and nothing in the repo tests or scores prompt effectiveness. Many "prompts" are really repackaged descriptions of other people's agent configs ("based on X repo, Apache-2.0, 50k stars"), which raises real attribution and license questions — the library folder ships separate LICENSE files for at least three different source projects (MIT, Apache-2.0), meaning license terms differ prompt-by-prompt and a user grabbing one file needs to check which one applies. With thousands of entries and an open PR-welcome model, there's no visible review process beyond link-checking, so quality and accuracy vary a lot between a one-off three-line prompt and a 50-line harness architect. A good chunk of the content is pegged to specific tool versions and features (CLI flags, file formats like SKILL.md) that will silently go stale as those tools evolve, with no deprecation or freshness signal in the index itself.
We read the README, the directory tree and a key source file in November 2026. Nobody pays to be listed. How we judge →