Use this to staff training and performance reviews for AI builders.
- Explains what a token is and that models have finite context.
- Reads provider pricing tables and understands input vs output billing.
- Writes concise prompts; closes irrelevant editor context; starts fresh chats between unrelated tasks.
- Uses a token counter or IDE meter when prompted.
- Budgets prompts with tiktoken,
count_tokens, or internal metrics. - Maintains lean
CLAUDE.md/AGENTS.md/ rules files; moves detail into skills. - Uses summarization and compaction deliberately; designs few-shot only when needed.
- Routes models by task risk; uses caching features where supported.
- Audits MCP footprint per project.
- Performs context breakdown reviews (what is always-on vs on-demand).
- Builds RAG pipelines with evaluation for chunking and retrieval quality.
- Ships custom MCP or CLI tools that return minimal sufficient context.
- Runs org-wide dashboards (provider consoles + internal traces); sets budgets and alerts.
- Trains others; defines lint rules for prompts and agent configs.
- Basic:
training/level-1-beginner/ - Advanced:
training/level-2-intermediate/+guidelines/ - Expert:
training/level-3-expert/+docs/04-mcp-guide.md+ external graph/RTK/token-optimizer pilots