Claude Mistakes System Tags for Injection Prompts
Das_Mohr · reddit · 2026-07-19
User feedback indicates that Claude Code mistakes certain tags provided by Anthropic as "injection prompts," specifically the Quick Answer and occasionally the longconversation tags.
Symptoms
- As soon as Quick Answer is enabled, Claude often proactively states it won't follow these "injected tags."
- It consistently treats these tags as user-injected commands, causing repeated interruptions to the conversation and making the overall experience highly unstable.
Core Issue
- The author believes this is a poor coordination between different instruction layers: the model seems unable to correctly distinguish between product system tags, internal control signals, and user input.
- The discussion focuses not on a single bug, but on the fragility of instruction layers / prompt hierarchy in real-world products.
More from coding & agent
- A roundup of AI agents and MCP resources, including how to evaluate agents — _jaydeepkarale · 2026-07-21
- Anthropic shares a masterclass on how it builds AI agents — _jaydeepkarale · 2026-07-21
- Anthropic masterclass spotlights how to build and observe AI agents — _jaydeepkarale · 2026-07-21
- A beginner guide to AI agents points readers to a Stanford webinar — _jaydeepkarale · 2026-07-21
- A full course shows how to build and deploy an AI agent with OpenAI and LangChain — _jaydeepkarale · 2026-07-21
- A practical guide on how to evaluate AI agents — _jaydeepkarale · 2026-07-21