Go to the line
Open the file NeuroShield names at the line it shows.
AI coding agents · Risk · CWE-1427 · OWASP LLM01
Your agent reads instructions you can't see
Some characters don't show up on screen. The text you read and the text the agent reads can differ.
Why it matters
Instruction files are read before every session. A line that looks harmless to you can carry text only the agent sees, and a rules file copied from somewhere else is the easiest place to hide it.
Where NeuroShield looks
| File | What it reads |
|---|---|
| Instruction files | CLAUDE.md, CLAUDE.local.md, AGENTS.md and each agent's rules files. |
| Skills, subagents, commands | Every file an agent loads as instructions. |
How to fix it
Retype or delete it
That line holds a character you cannot see.
Check the source
Check where the file came from before you use it again.
NeuroShield names the character and the line
It shows its code (U+…) and the line, so you can see what your editor hides.
Questions
Why don't I see it in my editor?
Those characters have no glyph, or they change how text is shown. The agent still reads them.
See your own agents
The check is free: how many agents you have and what they can do. Details and fixes come with Shields.