clade-data-handling — One-Pager
Manage the 200K context window, redact PII, and handle data retention for Claude integrations.
The Problem
Claude's 200K token context window is generous but not infinite. Long conversations overflow it and trigger invalid_request_error. Developers also need to handle sensitive data responsibly — sending raw PII (emails, phone numbers, SSNs, credit cards) to any external API is a compliance risk. And few teams understand Anthropic's default data retention policies or what they are responsible for on their end.
The Solution
This skill provides three production patterns: token counting with countTokens before every request to prevent overflow, a conversation trimming strategy that preserves the first message and recent turns while dropping the middle, and regex-based PII redaction for emails, phone numbers, SSNs, and credit card numbers. It also documents Anthropic's data retention defaults (no training on API data, zero retention on Enterprise).
Who / What / When
| Aspect | Details |
|---|---|
| Who | Developers building Claude applications that handle user data or long conversations |
| What | Context window management, conversation trimming, PII redaction, and data retention policy guidance |
| When | When your application processes user-generated content, handles multi-turn conversations, or must comply with data privacy requirements |
Key Features
- Token counting — Use
client.messages.countTokens()to measure input before sending, calculate available budget as200K - max_tokens - Conversation trimming — Keep the first message (key context) and the last 3 turns, drop the middle to fit within budget
- PII redaction — Regex patterns to strip emails, phone numbers, SSNs, and credit card numbers before sending to the API
- Data retention awareness — Anthropic does not train on API data by default; zero retention available on Enterprise plans; your responsibility to manage stored responses
Quick Start
See SKILL.md for full instructions.