Skill: token-efficiency

What I do

I optimise every AI interaction for maximum value per token: being explicit about intent, structuring information clearly, removing noise, and using iteration instead of perfection in one shot. I provide efficiency techniques that reduce costs identified by token-cost-estimation.

When to use me

• When asking complex questions or requesting implementations
• When dealing with large codebases (summarise, don't dump)
• When writing prompts that will be reused
• When you have limited token budget
• When token-cost-estimation identifies optimisation opportunities

Core principles

1. Explicit intent - State what you need, why, what success looks like
2. Structured information - Sections, bullets, clear formatting over prose
3. Cut noise - Remove unnecessary words and irrelevant context
4. Context efficiency - One good example beats ten vague descriptions
5. Iterate - Expect refinement, don't demand perfection first try

Efficiency Techniques

Prompt Structure (saves 10-20%)

Bad: "I need help with the authentication system, 
     it's not working properly and I've tried a 
     few things but nothing works..."

Good:
Goal: Fix auth token validation
Error: JWT expired check failing
Tried: Updated token library (no effect)
Need: Root cause + fix

Context Provision (saves 15-25%)

• Provide relevant code snippets, not entire files
• State assumptions explicitly
• Include error messages verbatim
• Reference specific line numbers

Efficient Patterns

Integration with token-cost-estimation

Pre-Session

1. Review token-cost-estimation breakdown
2. Identify high-cost phases
3. Apply efficiency techniques to reduce

During Session

• Use structured prompts throughout
• Provide focused context
• Iterate in small steps

Post-Session

• Compare actual vs estimated
• Identify which techniques helped
• Store learnings in memory-keeper

Quantitative Metrics

Track these to measure efficiency:

• Tokens per task completed
• First-attempt success rate
• Iteration count per task
• Context rebuild frequency

Anti-patterns to avoid

• ❌ Dumping entire files when snippet suffices
• ❌ Vague requests ("fix this")
• ❌ Expecting perfection on first try
• ❌ Repeating context unnecessarily
• ❌ Not learning from high-cost sessions

KB Reference

~/vaults/baphled/3. Resources/Knowledge Base/AI Development System/Skills/Workflow-Orchestration/Token Efficiency.md

Related skills

• token-cost-estimation - Quantifies costs, identifies savings
• pre-action - Clarify before prompting
• parallel-execution - Efficiency through parallelism
• scope-management - Scope affects token usage
• context-efficient-tools — Tool result filtering (complements prompt-level efficiency)