Behavior Intelligence · Scoped answers
The evidence for all six products preserves meaningful scope boundaries, but the mechanisms differ. Codex receives harness and repository context; the selected Claude evidence is tied to dated chat-product prompts; Claude Code, Gemini CLI, ChatGPT, and Cursor document layered project or user context. These conclusions should not be generalized to every model, plan, surface, or runtime configuration.
Six product answers
Keep instructions attached to the correct product and scope instead of assuming one context layer applies everywhere.
Codex's selected sources establish harness-supplied workspace context and repository guidance collected through AGENTS.md files.
The selected Claude evidence applies to dated claude.ai and mobile product prompts and explicitly separates that scope from Claude API behavior.
Claude Code documents managed, user, project, and local CLAUDE.md scopes plus separate repository-scoped auto memory.
Gemini CLI documents global, workspace, and just-in-time GEMINI.md context plus dynamic injection of enabled skills, subagents, and tools.
ChatGPT documents custom instructions across conversations, project-bounded instructions and files, separately controlled memory, and selectable model or mode context.
Cursor documents user rules across projects, version-controlled project and path-aware rules, repository memories, and separately selectable provider models.