home/roleplay/benchmark-prompts

Benchmark Prompts

GPTClaudeGemini··▲ 879 copies·updated 2026-07-14

benchmark-prompts.prompt

# Benchmark Prompts

These are the human-readable code-generation prompts used by the published
benchmark families.

| Prompt family | Prompt file | Used by codegen runs |
| --- | --- | --- |
| `basic-commerce` | [basic-commerce.md](basic-commerce.md) | `20260609_152615_16568`, `20260609_195509_38196`, `20260610_091239_64934`, `20260610_092511_66275` |
| `strict-production` | [strict-commerce.md](strict-commerce.md) | `20260610_141237_92652`, `20260610_202937_24877` |
| `strict-commerce-no-mcp` | [strict-commerce.md](strict-commerce.md) | `20260611_002935_41179` |

The raw run folders also preserve per-run prompt copies under
`benchmark/raw-results/.skill-codegen-runs/<run_id>/<arm>/prompt_*.txt`.
Those copies include the exact prompt text submitted during each run. The files
here exist so readers can inspect the benchmark tasks without digging through
the raw result tree.

when to use it

Community prompt sourced from the open-source GitHub repo AnamKwon/programming-as-theory-building-skill (MIT). A "Benchmark Prompts" style prompt — adapt the placeholders and specifics to your task. Imported as-is and not independently retested here, so check the output before relying on it.