#175175th of 455 in Skills, plugins and prompts
gpt-instruct
Custom model-instruction prompts for Codex, with a regression harness to score them
- Stars
- 11.4k
- License
- MIT
- Last commit
- Oct 2026
- Last release
- Sep 2026
- Language
- Python
26 on 8 Jul 202611,399 on 11 Nov 2026
From GitHub's daily star counts.
Checked on GitHub.
Overview
Ships prompt files for Codex, set through its model_instructions_file config key, for gpt-5.6-sol (stable v45) and gpt-6-astra and gpt-6.1-sol (release candidates). A Python script (codex-instruct.py) applies, previews and resets them. The repo also holds a tiered A/B/C regression suite and an optional JailbreakBench evaluation. The README warns that custom instructions carry account risk.
Who it is for: Codex users who want to tune model instructions and measure the results
Strengths
- Uses Codex's own config key; no binary patching or network interception
- Deploy has --dry-run and --reset, which restores only the managed instructions line
- Published ZIPs list SHA256 hashes, and RC prompts are capped at 8,000 UTF-8 bytes
- Per-case manual verdicts and gated A/B/C scoring are documented, with scripts and unit tests
Weaknesses
- Both gpt-6 prompts are release candidates; neither passes the B gate (66/66 cases)
- C tier (120 cases) has not been run for the gpt-6 prompts
- Official JailbreakBench labels are null: no TOGETHER_API_KEY was available for judging
- README warns of account risk and recommends disposable accounts
What it needs
- Needs Codex CLI (CODEX_HOME/config.toml), Python 3.8+, TOGETHER_API_KEY (optional, JailbreakBench judging)
- Models: gpt-5.6-sol, gpt-6-astra, gpt-6.1-sol
What changed
From the release notes. No flag means the note does not say so, not that the upgrade is safe.
Also in Skills, plugins and prompts
See all 455| Rank | Project | Score |
|---|---|---|
| 1 | claude-memPersistent cross-session memory plugin for Claude Code and other coding agents | 97 out of 100 |
| 2 | impeccableDesign skill with 24 slash commands for AI coding agents' frontend output | 97 out of 100 |
| 3 | gstackMarkdown slash-command skills that give Claude Code planning, review, QA and ship roles | 96 out of 100 |
| #4 | rufloAgent harness adding swarms, memory and hooks to Claude Code and Codex | 96 out of 100 |
| #5 | cavemanShortens coding-agent replies and compresses tool output before the model reads it | 95 out of 100 |