Why is Cursor ignoring my rules?

Short answer

Most ignored rules have a mechanical cause: the file is .md instead of .mdc, the rule is Apply Intelligently with a weak description, its globs do not match the file in context, or a Team or Project Rule overrides it. Rules also only apply to Agent (Chat), not to Tab or Bugbot. And even an Always Apply rule is guidance, not a hard block; use hooks for that.

Check these in order

Most ignored rules are not the model being stubborn. The rule never reached the conversation. Work through this list from top to bottom; each step rules out one cause.

# Check What goes wrong Fix
1 Extension A plain .md file in .cursor/rules/ is ignored, because it has no frontmatter Rename to .mdc, or move plain markdown to AGENTS.md
2 Where you expect it Rules apply only to Agent (Chat), not to Tab completion or Bugbot reviews For reviews, use .cursor/BUGBOT.md; Tab ignores rules
3 Rule type Apply Manually only works when you @-mention it Mention it, or change the type
4 Description Apply Intelligently with no or a vague description is rarely picked Describe when the rule applies, in the words of the task
5 Globs Apply to Specific Files only triggers when a matching file is in context Test the pattern against the actual paths; open or mention a matching file
6 alwaysApply plus globs With alwaysApply: true, globs and description are ignored Choose one: always on, or scoped with alwaysApply: false
7 Conflicts Team Rules beat Project Rules, which beat User Rules Look for a higher rule that says the opposite
8 Legacy file .cursorrules is legacy and will be deprecated Move it into .cursor/rules/ as an Always Apply rule

Cursor’s own help page gives the core of this list: check the rule type, give Apply Intelligently rules a description, and confirm the pattern matches the files you work on.

Wrong place, wrong tool

Two causes surprise people the most.

Bugbot. Teams write a careful set of rules and then see Bugbot approve code that breaks them. Bugbot does not read .cursor/rules/. It has its own files and rules, explained in How do Cursor Bugbot rules work.

Cloud Agents and other machines. A Cloud Agent reads the repository it checks out, so a rule that only exists in your working copy and was never committed is not there. Rule files in ~/.cursor/rules stay on your own machine. Account User Rules do sync, see How do you set Cursor rules that apply to every project.

When the rule loads but is still not followed

If the rule is clearly in context and Agent still goes its own way, the cause is usually in the text:

  1. Too long. Cursor recommends keeping a rule under 500 lines and splitting large ones. A long rule buries the one line that matters.
  2. Too vague. “Write clean code” gives Agent nothing to check. “Return errors with the class in src/lib/errors.ts” does.
  3. Contradictions. Two rules that disagree, or a rule that disagrees with the code Agent reads, make Agent pick one. Rules with the same filename in different folders both apply, because rules are identified by full path.
  4. Too much always-on. Every Always Apply rule competes for attention in every conversation. Move what is only sometimes relevant to globs, a description or a skill.
  5. Hidden in a reference. A file a rule points to is read only when Agent decides to open it. Anything that must always be in context belongs in the rule body itself.

The common mistake: treating a rule as a lock

“Always Apply” sounds like a guarantee. It means the rule is always in the conversation, not that the model always obeys it. Cursor’s rules documentation itself warns that AI guidance should not be the only security control. For things that must never happen, such as deleting a production database or reading a secrets file, use hooks. A beforeShellExecution hook can deny a shell command, and a beforeReadFile hook can block a file from being sent to the model. Hooks live in .cursor/hooks.json in the project or ~/.cursor/hooks.json for your own machine.

So the split is: rules for how you want work done, hooks for what is not allowed, and a linter or test for what can be checked automatically.

What this means for a team

A rule that silently does not load is worse than no rule: everyone assumes the agreement is enforced. When a teammate says “Cursor ignores our rules”, run the table above before rewriting the rule. And keep the rules few and current, so a new rule does not fight an old one. Writing rules that are easy to follow is covered in Cursor rules best practices.

Where Bombos fits

Rules break quietly because they are static files: they only work if the right file is in the right place in the right tool. Bombos is a plugin that gives every AI tool your team uses, Cursor included, one shared memory with team knowledge and rules learned from corrections, and it shows what each answer rests on, so you can see which knowledge was used. It runs in the cloud or on your own server, and one developer can try it first next to their existing rules.

Related questions

I set alwaysApply: true and added globs. Why does the glob do nothing?

When alwaysApply is true, Cursor ignores the globs and description. The rule goes into every chat session. To scope a rule to files, set alwaysApply to false and keep the globs.

Does Cursor still read .cursorrules?

Cursor calls .cursorrules legacy and says it will be deprecated. Move the content into a new rule in .cursor/rules, set it to Always Apply, and delete the old file.

Is CLAUDE.md applied in Cursor?

Yes. Cursor reads CLAUDE.md like AGENTS.md and always applies it, regardless of alwaysApply settings.

How do I make sure the agent can never run a certain command?

Use a hook, not a rule. A beforeShellExecution hook in .cursor/hooks.json can deny a command; a rule only asks the model not to run it.

Sources

  1. Cursor docs: Rules
  2. Cursor help: Rules
  3. Cursor docs: Hooks
  4. Cursor docs: Bugbot
The first half hour · free

What does your AI forget?

  1. 01You tell us which AI you use and where it keeps starting over.
  2. 02We show what a shared memory changes there, on your own examples.
  3. 03You keep the plan, with or without Bombos.
Piet Baudoin
Piet BaudoinYou talk to Piet, from the Bombos teamDutch team from Utrecht and Nijmegen · data in Europe
We reply the same working day.