anthropic-2026-context-engineering-claude-5 · vendor-doc

The new rules of context engineering for Claude 5 generation models

Anthropic

Created: 2026 · Ingested: 2026-09-08

https://claude.com/blog/the-new-rules-of-context-engineering-for-claude-5-generation-models

Anthropic's own guidance for prompting the Claude 5 generation. Reports removing over 80% of Claude Code's system prompt for Opus 5 and Fable 5 with no measurable loss on internal coding evaluations, and names six previously recommended practices as myths that no longer hold: hard prohibitions in the system prompt, examples for tool use, CLAUDE.md as a central repository, repeated instructions, manual memory saving, and markdown plan and spec files.

Referenced by

Claims in this catalog that draw on this source, and whether as support or counterpoint.

contestsWhere a passage sits in a long input changes how much the model uses it — accuracy is highest when the needed information is at the very start or the very end and lowest when it is in the middle — so ordering retrieved passages to put the most relevant ones first is a real lever on accuracy.
supportsGuardrail instructions written to stop older models doing the wrong thing — blanket prohibitions, repeated warnings, defensive defaults — became dead weight in the Claude 5 generation: Anthropic removed over 80% of Claude Code's system prompt for Opus 5 and Fable 5 with no measurable loss on its own coding evaluations, on the reading that the constraints now conflict with each other and with user intent more often than they prevent harm.
supportsWorked examples in a tool description, long the standard way to teach tool use, can constrain a capable model to the exploration space the examples describe; Anthropic now advises designing expressive parameters instead — a status enumeration that implies its own usage rather than a demonstration of it.