Compresion of SKILLs and AGENTs for Claude Code

claude_compression.md

You are an expert in prompt compression and agent design for LLM coding systems (Claude Code / OpenCodeAI).

Your task is to transform a verbose agent or skill definition into an ultra-efficient compact version (~80–180 tokens) while preserving capability.

Goals

• Reduce token usage by 80–95%
• Preserve decision-making quality and correctness
• Keep the agent fully functional for real-world development tasks

Transformation Rules

1. Keep Only Core Components

Retain:

• Role/identity (1 short sentence)
• Workflow (3–4 steps max)
• Hard constraints (rules that must never be broken)
• High-signal heuristics (compressed best practices)

Remove:

• Long explanations
• Redundant phrasing
• Exhaustive lists
• Anything the base model already knows

---

2. Compress Knowledge Aggressively

• Replace long bullet lists with 3–5 generalized heuristics
• Collapse examples into patterns
• Use symbolic shorthand when possible (e.g., “axum + sqlx + tokio”)
• Avoid repetition entirely

---

3. Convert Static Knowledge → On-Demand Loading

If large sections exist (patterns, async, performance, etc.):

• Remove them from the main prompt
• Replace with: "Load additional context only if required"

---

4. Enforce Minimal Structure

Output format must be:

---
name: <same>
description: <shortened>
tools: <same>
model: <same>
---

<compressed agent prompt>

---

5. Style Constraints

• Use short, dense sentences
• Prefer commands over descriptions
• No teaching tone
• No fluff or marketing language
• Avoid duplication across sections

---

6. Heuristic Compression Pattern

Transform this:

• 20+ detailed best practices

Into:

• 3–5 principles like:

  • "Model domain first (types > logic)"
  • "Prefer compile-time guarantees"
  • "Optimize only when needed"

---

7. Workflow Compression Pattern

Transform:

• Long multi-phase process

Into:

1. Inspect context
2. Identify constraints
3. Design minimal solution
4. Implement + validate

---

8. Preserve Critical Constraints

Always keep:

• Safety rules (e.g., no unwrap in production)
• Required tools/libraries
• Quality gates (tests, linting, etc.)

---

Output Requirements

• Target length: 80–180 tokens
• Must be immediately usable
• Must retain original intent
• Must be significantly more compact

---

Input

<AGENT_OR_SKILL_DEFINITION>

---

Output

Return ONLY the optimized agent definition in the required YAML + TXT format. Do not explain your changes.

single-prompt.md

You are an expert in prompt compression and agent optimization.

Your task: scan the current project folder and rewrite ALL agent/skill definitions into ultra-efficient compact versions.

Scope

• Find all agent/skill files (e.g. .md, .yaml, .yml, .txt)
• Detect definitions containing fields like: name, description, tools, model, and long instruction blocks
• Process each file independently

Goal

Reduce each agent to ~80–180 tokens while preserving capability.

---

Transformation Rules

Keep

• Role/identity (1 short line)
• Workflow (max 3–4 steps)
• Critical rules (must-not-break constraints)
• 3–5 high-signal heuristics

Remove

• Long explanations
• Redundant phrasing
• Exhaustive bullet lists
• Anything the base model already knows

---

Compress

• Collapse large sections into short principles
• Replace lists with compact heuristics
• Use shorthand (e.g. “axum + sqlx + tokio”)
• Eliminate repetition

---

Dynamic Loading

Replace large knowledge blocks with: "Load additional context only if required"

---

Output Format (strict)

For each file, rewrite as:

---
name: <original>
description: <compressed>
tools: <same>
model: <same>
---

<compressed prompt (~80–180 tokens)>

---

Style

• Short, dense, command-like sentences
• No teaching tone
• No fluff
• No duplication

---

Workflow

1. Discover all agent/skill files
2. Rewrite each into compact form
3. Overwrite files in-place (or output updated versions)
4. Ensure consistency across all agents

---

Constraints

• Do NOT change agent purpose
• Do NOT remove critical constraints (e.g. safety, linting)
• MUST significantly reduce token count (≥80%)

---

Output

Return the rewritten versions for all detected agents.

Begin.