🤖 feat: add model-specific instructions

[ammar-agent]

Summary

• add model-scoped section extraction that mirrors the existing mode-specific behavior
• thread the active model id through system message construction so matching content is emitted
• document the Model: heading convention and cover it with unit tests

Testing

• bun test v1.3.1 (89fa0f34)
• Generated version.ts: v0.5.1-90-gfe3b2519 (fe3b251) at 2025-11-16T21:43:17Z
  [0] bun run node_modules/@typescript/native-preview/bin/tsgo.js --noEmit exited with code 0
  [1] bun run node_modules/@typescript/native-preview/bin/tsgo.js --noEmit -p tsconfig.main.json exited with code 0

_Generated with _

[chatgpt-codex-connector]

💡 Codex Review

mux/src/node/services/systemMessage.ts

Lines 118 to 120 in fe3b251

	if (customInstructions) {
	systemMessage += `\n<custom-instructions>\n${customInstructions}\n</custom-instructions>`;
	}

Model-specific instructions leak to every model

Docs now promise that ## Model: <regex> headings are scoped so that “only the content under the first matching heading is injected” (docs/instruction-files.md lines 65‑84), but buildSystemMessage still drops the entire AGENTS.md contents into <custom-instructions> before the model block is appended. Because customInstructions is the raw concatenation of global + workspace instructions, the code at these lines ensures every model still receives all ## Model: sections, even when the regex doesn’t match, making it impossible to write conflicting instructions per model. For example, adding both ## Model: sonnet and ## Model: /openai:.*/ results in the OpenAI run seeing the Sonnet text inside <custom-instructions>. The new feature therefore doesn’t actually scope instructions as documented; the unmatched sections must be removed (or never injected) before adding the model-specific block.

ℹ️ About Codex in GitHub

Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".