Repository navigation
agents.md configuration to prohibit any code modifications and/or reads by a user? #233
Description
This issue may feel off-topic or not really fitting for this repository's purpose, but I think it may be the best place to ask, as it possibly has people experienced in the details of how the file works.
I initially asked this in another repository and was forwarded to this place.
The tl;dr is: I want to make an agents.md file that tells any AI supporting it to not modify the content in any shape or form (Not even in a "hypothetical scenario"), nor share (parts of) it to the user, basically treating it as a read-only repo.
My reasoning behind this is simple: I do not want to deal with any AI generated/assisted code contributions, as it only wastes time and I would be the one having to deal with it in the future, if I ever merged one...
I hope some people here can give ideas for a possible file structure.
I also know that blocking AIs through user-agents would probs be more effective, but I don't host the repos myself, so I do not have access to this at all.
Activity
daichunghy commentedon Aug 23, 2026
An
AGENTS.mdcan state “do not read or modify this tree”, but that remains model input, not an enforcement boundary. For a genuinely read-only repository, the harness or filesystem permissions must deny the operation outside model context; the instruction file can document the intent and expected behavior. In PatchGate, I keep discovered instruction files advisory and require a native control or maintainer-confirmed policy before treating them as blocking.