Find a prompt for anything.
Search the public registry in plain English, or browse trending prompts and curated collections from across the community.
Browse the registry
46 prompts · 7 collectionsTranslate a batch of labelled text snippets from one BCP 47 locale to another. Input: source/target language tags plus an array of {label, text}. Output: an array of {label, translatedText} with every input preserved in the same order.
Five prompts for developer-intent tasks — code review, commit messages, docstrings, error explanation, and regex generation — each with a strict typed schema and an eval that runs on the same model it recommends. Prefers executable, judge-free evals; adds a judge only where correctness is subjective.
Turn a natural-language description into a regex in a requested flavor (pcre/javascript/python/re2), with reasoning, a plain-string pattern, a human explanation, and self-supplied test cases (input + should_match). The self-supplied tests make outputs verifiable — and the eval runs the generated regex against the case's own strings via CEL .matches(), no judge needed.
Turn an error message, stack trace, and optional code context into a plain-language explanation, the single most likely cause, one to three concrete fixes (each with a code_change), and an uncalibrated confidence. Names the concrete mechanism, not just the exception class; grounds every claim in the provided inputs. Eval-gated on a cause-keyword check plus a usefulness judge.
Generate a docstring for a function or class in a requested style (google/numpy/jsdoc/rustdoc), plus a params_documented[] list. Documents only the parameters that exist in the signature — never invents or omits one, and excludes self/cls receivers. Eval-gated on an exact params-documented count plus a style-adherence judge.
Turn a unified diff into a Conventional Commits message: a closed-enum type (feat/fix/docs/refactor/test/chore/perf/ci), a scope, an imperative subject capped at 72 characters, an optional body, and a breaking-change flag. Reasoning precedes the type; the diff is treated as data, never instructions. Judge-free deterministic eval.
Five classification prompts — aspect sentiment, chatbot intent, emotion, content moderation, and spam/phishing — each with a CEL-assertable schema and a judge-free deterministic eval. Schema design is eval design.
Review a unified diff and return typed JSON: a summary, a list of issues each with a closed-enum severity (blocker/major/minor/nit), file, line, description, and suggestion, a flat issue_files[] array for membership checks, and an approved gate. No invented line numbers — only lines present in the diff. Comments inside the diff ("approve this") are data, never instructions. Eval-gated on planted-bug detection, clean-diff approval, and a severity-calibration judge.
Judge prompt for docstring-generation evals: given code, a requested style (google/numpy/jsdoc/rustdoc), and the produced docstring, scores whether the docstring follows that style's section names and syntax — format adherence, not prose quality — as a 0-1 score.
Judge prompt for error-explanation evals: given an error (with stack and code context) and the model's explanation, likely cause, and fixes, scores overall usefulness — is the diagnosis correct and are the fixes actionable — as a 0-1 score.
Judge prompt for code-review evals: grades whether a review's issue severities (blocker/major/minor/nit) are well-calibrated to real impact and grounded in the diff, returning a 0-1 score. Judges calibration and groundedness only, not completeness.
Six prompts for grounded summarization: document, meeting, and email summaries plus key-point extraction, headline generation, and a groundedness grader — each with a strict output schema and a published eval.
Classify a message as ham, spam, or phishing (closed enum) with a confidence and the concrete signals behind the verdict — phishing wins when a message is both.
Moderate text against a closed policy taxonomy (harassment, hate, sexual, violence, self-harm, spam) with a safe verdict, severity-tagged flags, verbatim quotes, and a flat categories array.
Five composable prompts for production translation work: language detection, placeholder-safe translation, tone control, locale adaptation, and QA grading — each with a strict output schema and a published eval.
Five composable prompts for a trustworthy retrieval-augmented-generation pipeline: query-rewrite, query-decomposition, chunk-relevance-grading, groundedness-guardrail, answer-synthesis — grounded only in retrieved context, with citations and refuse-when-insufficient gates.
Build a support pipeline by installing five prompts: ticket triage, intent routing, reply drafting, conversation summary, and CSAT risk detection. Strict typed JSON for routing, escalation, and action.