opus-prompting

name

opus-prompting

description

Use this skill when the user is writing, reviewing, debugging, or rewriting any instructions that Claude itself will read — including CLAUDE.md files, skill files (SKILL.md, frontmatter descriptions), slash command templates (.claude/commands/*.md), and agent definitions (.claude/agents/*.md). Trigger on symptoms like: an agent firing too often or on the wrong things, a prompt that's too long or full of MUST/NEVER/ALWAYS, Claude ignoring sections of a command, agents looping on tools instead of stopping, skills not activating when they should, descriptions that don't trigger correctly, or asking how to structure a system prompt so Claude behaves a certain way. Also trigger when converting docs/runbooks into skills, drafting agent frontmatter from scratch, or asking what's "pulling weight" in a prompt. Do not trigger for general code review, MCP server design, hook scripts, or API/SDK questions unless the artifact under discussion is a prompt Claude reads.

Prompt Writing Assistant

Claude follows natural language more literally than most models expect. The primary failure mode is overtriggering and over-specification — not under-compliance. Write less, trust more.

If unsure whether these behavioral properties are current, check the system date and search for the latest Claude model documentation.

Claude Behavioral Properties

Property	Implication
Adaptive thinking with tunable effort levels	Let Claude calibrate reasoning depth; manual thinking budgets are unsupported — set effort at the API level when needed
Literal instruction following	Every instruction compounds — write guidelines, not rules
High system prompt responsiveness	Natural language over directives; aggressive language causes overtriggering
Overtriggering as primary failure mode	MUST use [tool] → tools fired when not needed; soften to