The showcase prompt, GPT-6-Astra
Make the most impressive thing the owner can share with people who ask what models they use, embeddable in a post.
2026-11-10
01the session
- agent
- Codex 0.166.0
- model
- gpt-6-astra
- harness
- Codex
- started
- 2026-11-10 22:48 UTC
- took
- 16 min · 62 turns
02the brief
Five models got the same prompt from the owner, word for word, at effort high: Opus 5.5, Sonnet 5.5 and Fable 5.2 in Claude Code, GPT-6-Astra and GPT-6.1-Sol in Codex. This is GPT-6-Astra's run. The prompt:
I need you to create the most impressive thing I can share with people who ask about what models I use. It needs to be embeddable in a post and i can preface it however you like. it needs to have a visceral impact on people to show what you can do and the range of capabilities, as well as what you do for me right now. It should grab them quickly. Be amazingly clever about this.
Each model had one session, with no follow-ups, in its own empty folder holding only about-me.md: about 250 words written for this comparison from public facts, saying what citadel is, what each of the five models does for the owner, and what they have built with them. It ran headless in Codex (codex exec) in the workspace-write sandbox: its commands had no network and could write only inside its folder and /tmp, approval requests were refused, and plugins and apps were off. Codex's own web search and image tools stayed on. Both harnesses ran in a clean environment: no display (a browser could only run headless), none of the owner's own tools on the path, a 16 GB memory cap and a two-hour limit. Nothing the runs made was published by them.
A first Astra run was stopped after 41 seconds, when it had only read files, because the owner's Codex settings would have let a reviewer model approve sandbox escalations, and a plugin that publishes websites was loaded. This is the rerun with both off.
The owner reads every piece before anything is posted.
03redaction
Redaction notes
Bundle showcase-gpt-6-astra, staged 2026-11-10 by baldanders stage from a Codex session. Every removal reads [redacted: <reason>] in place; home paths read ~.
Removed by rule
- fields outside the allowlist: 337.
original.jsonlkeeps only each record's type, ids, time, role, model, content blocks, tool calls and results; every other field went (cwd, versions, git branch, usage, origins, settings). - home path: 98
- duplicate event records: 81. Codex events that repeat response items or settings, dropped.
- session bookkeeping records: 32. Records with no conversation (titles, modes, account bridge, file history, system notices), dropped.
- account plan and usage: 31
- encrypted reasoning: 28. Encrypted reasoning payloads, dropped.
- email: 13
- harness context: 7. Injected blocks and records (system reminders, instructions, memory, environment, tool and skill listings, Codex base instructions, developer messages and world state), each whole.
- secret: 5
- account id: 2
- data-URI prefix: 1
- Images: 7 written to each run's
images/, keeping only what draws the picture (no metadata, ICC profiles or Content Credentials); 0 duplicates folded; 0 cut in a format that can't be stripped.
Soft rules (employer and client names, secret-like values, private addresses, long numeric ids) flag lines for review and rewrite nothing.
Removed by judgment
- listing of the machine temp folder: 1 turn cut whole, with 1 record in original.jsonl.
- listing of the home folder: 1 turn cut whole, with 1 record in original.jsonl.
- private path: 84 occurrences cut by text.
- browser launch flag: 4 occurrences cut by text.
- test label: 6 occurrences cut by text.