prompt-writer

To write reliable system prompts, ground instructions in verified target content and assemble XML-delimited message architecture with explicit scope, diverse examples, and locked output shapes to eliminate guesswork and drift.

Auditing existing prompts requires an anti-pattern-first flow that checks for negative-only instructions, vague scope, prefill patterns, unsafe formatting, and output-shape ambiguity to identify and resolve common prompt-writing failures.

Structure XML tags by splitting system versus user messages into consistent tag-wrapped blocks, adding clear scope definitions, recency reminders, and output shape locking to create high-signal structured prompts for Claude execution.

Yes, this approach applies to authoring and editing CLAUDE.md content, .claude/rules files, SKILL.md bodies, slash-command bodies, and subagent briefings that require structured instruction formatting for other Claude agents to execute.

Inconsistent output often stems from under-specification, vague scope, or output-shape ambiguity in the prompt, which can be resolved by grounding instructions in actual content and enforcing XML-delimited architecture with locked output shapes.

XML-structured prompts are unnecessary for simple, single-turn tasks that do not require strict output shape locking, explicit scope enforcement, or anti-pattern auditing to achieve reliable Claude execution.