skills.rest
ZipperCode
ZipperCodeCOMMUNITY
Explore all skills by this author →
GitHub

claude-code-api-compaction

Implements gateway tolerance for Claude Code context-management fields in Anthropic API requests.

4|1|Updated Apr 29, 2026
Search Tags:#context management#claude code#api gateway#compaction#prompt caching#anthropic api#proxy compatibility
One-click install
npx skills add https://github.com/ZipperCode/lingma2api --skill claude-code-api-compaction-zippercode
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: claude-code-api-compaction
Source: https://github.com/ZipperCode/lingma2api/tree/main/.claude/skills/claude-code-api-compaction
Command: npx skills add https://github.com/ZipperCode/lingma2api --skill claude-code-api-compaction-zippercode

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve? Claude Code clients send context-management fields (context_management, cache_edits, cache_reference, cache_control) that cause Anthropic-compatible gateways to reject /v1/messages requests with 400 errors when the gateway does not recognize these extension fields. ## Core Features & Use Cases - Field Tolerance Rules: Defines minimum handling for context_management edits, cache_edits blocks, cache_reference in tool_result, cache_control markers, and context-management-* betas so they can be ignored or forwarded without failing requests. - Wire Format Reference: Documents the exact JSON shapes of clear_tool_uses_20250919, clear_thinking_20251015, cache_control, cache_reference, and cache_edits payloads. - Compact Request Handling: Treats non-streaming compaction summary requests (stream:false, no tools, small models) as ordinary /v1/messages calls. - Use Case: When building an OpenAI-to-Anthropic proxy, apply this Skill so Claude Code sessions with prompt caching and context compaction continue working even though the upstream does not implement real prompt caching. ## Quick Start Use this skill to make my Anthropic gateway tolerate Claude Code context_management and cache fields without returning 400 errors.

Frequently Asked Questions about claude-code-api-compaction

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I handle Claude Code context_management fields in an Anthropic gateway?▼

Accept the context_management object and its edits array (clear_tool_uses_20250919, clear_thinking_20251015) by ignoring or forwarding them. Do not return a 400 error even if your upstream does not implement these clearing strategies.

What is the cache_edits block in Claude Code requests?▼

cache_edits is a content block in user messages containing edit operations like deleting a cache_reference. Gateways without real prompt caching should ignore this block and continue processing the text or tool_result content in the same message.

Does my proxy need real prompt caching to support Claude Code?▼

No. Cache fields are performance optimizations and must not block compatibility. You can ignore cache_control and cache_reference, and return usage with cache_read_input_tokens and cache_creation_input_tokens set to 0.

Why does /v1/messages return 400 errors with Claude Code clients?▼

The gateway likely rejects unknown extension fields such as context_management, cache_edits, or context-management-* betas. Fix this by tolerating these fields instead of strict-validating the request schema.

How should compaction requests from Claude Code be processed?▼

Treat them as ordinary non-streaming /v1/messages requests. They typically use stream:false, no tools, and a small model like Haiku, so no special handling is required beyond normal request forwarding.

Related Skills

Explore More Software Engineering Skills→
AI-Staffing-Solution-Consultants-LLCAI-Staffing-Solution-Consultants-LLC

agency-ai-gateway

Provides Cloudflare AI Gateway context for proxying, routing, and monitoring LLM API traffic.

Official
Basic
smorky850612smorky850612
37

aura-guard

Monitor Claude context usage and trigger auto-compact at a configurable threshold.

Community
Advanced
latestaiagentslatestaiagents
5

prompt-caching-patterns

Implement provider-level, response, semantic, and template caching for LLM applications.

Official
Advanced
somachaksomachak

claude-api

Integrate Claude API with prompt caching and multi-surface routing.

Community
Advanced