writing-claude-directives_skill

Overview

This skill helps authors write precise directives for Claude-based agents, system prompts, and skill descriptions. It focuses on token efficiency, motivation-driven compliance, and discovery-friendly metadata so directives are effective and concise. Use it to translate project rules and workflows into prompts that Claude follows reliably.

How this skill works

The skill inspects intended behaviors, high-stakes constraints, and the target model (Claude 4.x) then recommends framing, placement, and structure patterns that maximize compliance. It prescribes concise motivations, boundary repetition, XML/structured formats for preservation, and progressive disclosure to limit token load. It also generates discovery-ready descriptions and naming guidance for skills.

When to use it

• Writing system prompts, agent prompts, or skill/feature directives for Claude
• Converting team workflows, tests, or verification steps into machine-friendly instructions
• Optimizing long prompts for token cost and attention (placing critical rules at boundaries)
• Creating discovery-metadata (skill descriptions, keywords) so Claude finds the skill
• Enforcing compliance for high-risk operations (secrets, CI gates, release steps)

Best practices

• Lead with context and motivation so Claude generalizes the rule rather than just obeying an imperative
• Prefer positive framing: tell Claude what to do instead of what to avoid
• Keep frequently loaded directives under ~200 words and main skill docs concise with progressive disclosure
• Place critical instructions at the start or end of prompts and repeat high-stakes rules in different phrasings
• Use structured formats (XML or explicit tags) for multi-part directives and verification gates
• Prune to ~150 instructions; prefer one clear default approach and an explicit escape hatch

Example use cases

• Authoring a Claude skill that must never commit secrets: craft motivation + blocking rule + verification step
• Turning a TDD workflow into a prompt: numbered steps with commit gates and rollback behavior
• Optimizing a long task prompt by moving nonessential details to linked reference files
• Creating a skill description that triggers discovery for flaky tests or race conditions
• Converting policy into machine-friendly XML tags for consistent parsing and output formatting