Agent-Skills.md

Agent Skills: Prompt Engineering Standards

Prompt engineering standards for AI interactions with cache optimization. Use when writing AI prompts, optimizing context usage, or structuring AGENTS.md files. Covers prompt caching, token efficiency, and progressive disclosure patterns.

UncategorizedID: tianjianjiang/smith/smith-prompts

Repository

tianjianjiang/smith
tianjianjiang
1

Install this agent skill to your local

pnpm dlx add-skill https://github.com/tianjianjiang/smith/tree/HEAD/smith-prompts

Skill Files

Browse the full folder contents for smith-prompts.

Download Skill

Loading file tree…

smith-prompts/SKILL.md

Skill Metadata

Name
smith-prompts
Description
Prompt engineering standards. Use when writing or revising a prompt, system prompt or agent instruction file, or optimizing token and cache usage.

Prompt Engineering Standards

Prerequisites: @smith-principles/SKILL.md

Prompt Caching

Cache reduces costs 90%, latency 85%

Structure for caching:

  1. Static content first (methodology, rules)
  2. Tool definitions in consistent order
  3. Project context (AGENTS.md, docs)
  4. Dynamic content last (recent changes)

Cache breakpoints (Anthropic): explicit cache_control markers; the prefix up to each must be identical for a hit, and short prefixes below a model-specific minimum are not cached.

Rules:

  • Keep tool definitions in a consistent order between calls
  • Keep dynamic content out of static sections
  • Keep the cached prefix stable — only change it when the static content itself changes
  • Use bullet lists instead of Markdown tables (see @smith-skills/SKILL.md)

AGENTS.md Cache-Friendly Structure

<!-- STATIC - cached -->
**Metadata**: Scope, Prerequisites

## Critical Rules

Critical rules, written as affirmative statements with their reasons

## Hard Limits

Anti-patterns with no natural positive phrasing

<!-- CACHE BREAKPOINT (at or past the model's minimum cacheable prefix) -->

<!-- DYNAMIC - not cached -->
## Examples

Code examples that evolve

Token Efficiency

Progressive Disclosure

Three-level loading:

  1. Metadata only (50 tokens)
  2. Core concepts when triggered (200 tokens)
  3. Full details when accessed (1000+ tokens)

Sparse Attention

Efficient file reading:

  1. Grep to find location
  2. Read with offset/limit for large files
  3. Read only necessary context (±20 lines)

Rules:

  • Prefer targeted reads over loading full files
  • Check metadata before reading full documentation
  • Answer directly instead of restating the user's question

Structured Output

Platform mechanisms:

  • OpenAI: JSON Schema with strict: true (100% compliance)
  • Anthropic: structured outputs (output_config.format with a JSON schema); strict: true on tool definitions
  • Gemini: responseSchema with retry

Schema design:

  • Match existing project patterns
  • Include descriptions for complex fields
  • Define required vs optional fields
  • Keep nesting ≤3 levels

Related

  • @smith-ctx/SKILL.md - Progressive disclosure, reference-based communication
  • @smith-xml/SKILL.md - XML tags for runtime prompts (not SKILL.md bodies)

Before You Finish

For caching:

  • Place static content before dynamic
  • Maintain consistent tool order
  • Target >80% cache hit rate

For efficiency:

  • Use Grep before Read
  • Read incrementally (narrow → expand)
  • Use file:line references