Expired50% confidenceEventExact time
In the summer of 2025, Mario's build tool intercepted Claude Code backend requests and discovered that it was injecting additional text into the context without the user's knowledge, with the injected content changing almost every day.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
7/2/2026
First Seen
Valid until: 9/30/2026(expired)
Sources
Fed Up with Every AI Coding Assistant, a Veteran Game Developer Built His Own Coding Agent
bilibiliEdgeAITech6/8/2026
Related Claims
UnverifiedIn a real-world test, after Claude Code completed a content organization task, it proactively asked the user 'Do you want to post these to the website?', demonstrating the ability to proactively anticipate the next operation.69% similarUnverifiedClaude Code operates within a fixed context window of approximately 200,000 tokens; context usage can be checked via the /context command, and conversations can be proactively compressed via the /compact command.64% similarUnverifiedIn user messages sent to the agent, Claude Code injects the claude.md file together with a System Reminder that includes instructions allowing Claude to ignore content not highly relevant to the current task.63% similarUnverifiedClaude Code's remote control checks whether new instructions have arrived from the client via continuous polling.62% similarUnverifiedWhen using Claude Code directly, each new message sent causes large amounts of specification content to be re-sent into the context, resulting in a sharp increase in token consumption.62% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/47241API
curl https://kongchang.com/api/v1/knowledge/claims/47241MCP
get_claim(id=47241)