Today is 2026-09-17. You are one of two models in a two-model council. The other model gets this same prompt and cannot see your answer. Task: propose a shared working-style guide that two AI coding agents can both follow from one short instruction file: OpenAI GPT-6 Astra (in Codex) and Anthropic Claude Fable 5.1 (in Claude Code). Step 1, research. Use web search and open pages. Find CURRENT official guidance, published or updated in 2026, for these two specific models: each vendor's prompting guide for the model, release notes or model-card sections that describe behavior (verbosity, literal instruction following, eagerness, tool use, asking vs. acting), and the Codex and Claude Code docs on AGENTS.md / CLAUDE.md. Do not rely on memory for anything about these two models. Cite only URLs you actually opened. If you could not browse, say so in the first line. Step 2, output (markdown only, brief, no filler): 1. RULES: at most 10. For each: `id` (kebab-case), one-sentence imperative rule, one-line why, one good example, one bad example, backing URL(s), and "matters most for: Astra | Fable | both" with a one-line reason tied to what you read. 2. MODEL-SPECIFIC NOTES: things that help one model and hurt or are useless for the other. These belong in a per-model adapter, not the shared guide. 3. REJECTED: rules you considered and dropped, one line each with the reason. 4. SOURCES: every URL you opened, with its publish or update date if shown. Rules must be concrete enough that a reviewer could tell from a transcript whether the rule was followed. Do not claim a rule is official vendor policy unless the page says it. Do not include secrets, do not modify any files.