Codex Prompt Corpus

Summary

This corpus preserves high-quality user-authored Codex prompts and automation prompts as reusable prompt engineering material.

Selection Rules

  • EXTRACTED: Include user prompts and automation prompts that are substantive, clean, reusable, and not mostly code/log output.
  • EXTRACTED: Exclude short prompts, duplicates, mojibake/noise, pasted stack traces, server logs, source-code dumps, and secret-like/private material.
  • EXTRACTED: Public pages may include full selected prompt text, but only after safety filtering.

Current Counts

  • Selected prompts: 353
  • Rejected candidates: 1424
  • Manifest: raw/codex-prompts-public/manifest.json

Categories

automation

coding-agent-workflow

general-high-quality-prompt

personal-rules

prompt-engineering-pattern

research-workflow

wiki-ingest

Sources

  • EXTRACTED: Raw manifest and capture metadata are stored under raw/codex-prompts-public/.
  • EXTRACTED: Source material comes from local Codex session JSONL files and local Codex automation TOML files.

Counterpoints and Gaps

  • AMBIGUOUS: Local session encodings can be messy; future runs should continue filtering mojibake, terminal paste, and sensitive personal data aggressively.
  • INFERRED: Full-text prompt preservation is useful only for selected reusable prompts; rejected candidates should remain raw audit metadata, not public pages.