Explore Prompts

Take a prompt and inject 10 practical guardrails: forbidden content notes, schema validation, refusal policy, and fallbacks. Provide both a short and long version. Output as two Markdown blocks.

Given a single messy prompt, split it into System, Developer, and User roles. Explain why each instruction belongs where. Output: role, instruction, reason, risk_if_misplaced.

Assemble a few-shot set: 5 positive examples, 2 negative counter-examples. Each example contains: input, rationale (1–2 lines), correct output. Return JSONL, then propose an ordering strategy to minimize confusion and token cost.

Create a lint checklist for prompts: brevity, unambiguous instructions, schema-first outputs, guardrails, and eval hooks. Return a checklist table with yes/no and examples. Add 5 before→after rewrites that follow the guide.

Build a loop that collects model errors/failures and proposes prompt edits. Steps: (1) bucket failures, (2) extract missing instructions, (3) propose deltas, (4) generate 3 patched prompts, (5) quick re-test table. Output as YAML with sections: failure_buckets, root_causes, proposed_deltas, candidates[], retest_plan.

Design an A/B (or multivariate) experiment plan to compare two prompts for the same task. Include: hypotheses, metrics (precision/recall or qualitative rubric), sample size heuristic, randomization method, and decision rule. Output a single CSV with columns: metric, definition, calculation, target, min_effect_size, stop_rule.

Refactor this prompt into a testable brief with: goal, constraints, input schema, output schema, and acceptance tests. Provide 3 variants: (A) speed-optimized, (B) accuracy-optimized, (C) cost-optimized. Return JSONL: one line per variant with keys goal, constraints[], input_schema, output_schema, tests[].

Outline financing/leasing options for machinery upgrades in Lanzhou for SMEs vs larger enterprises. Return a comparison table: option, typical term, upfront %, monthly example, collateral note, pros/cons. Add a cash-flow stress test and covenant watchlist.

Create a B2B outreach plan targeting Lanzhou industrial buyers via WeChat and Douyin. Return: ICP profile, content pillars (case studies, process videos), posting cadence, live-demo format, lead capture flow, and a 4-week content calendar.

Draft a winter maintenance calendar for outdoor equipment in Lanzhou (approx. −15°C to 5°C). List monthly checks: lubrication, seals, heating jackets, battery care, snow/ice handling, and emergency parts bin. Return as a Gantt-like table with responsibility roles.

Provide a stormwater inspection and discharge monitoring template for a site near the Yellow River. Return a table: location, date/time, weather, turbidity check, sheen/odor note, sample taken(Y/N), corrective action. Add a quarterly housekeeping plan for inlets/drains.

Create an audit pack for power quality in a Lanzhou workshop: voltage dips, harmonics, power factor. Return: instrument list, sampling plan, acceptance thresholds (placeholder ranges), root-cause tree, and a corrective-action log template.