Repository navigation
[FEATURE]: Support path-scoped rules similar to Cursor globs #37463
Description
Feature hasn't been suggested before.
- I have verified this feature I'm about to request hasn't been suggested before.
Describe the enhancement you want to request
OpenCode currently provides:
AGENTS.mdfor always-loaded instructions- Skills for semantic/on-demand behavior loading
However, there is currently no equivalent to Cursor's path-scoped rules:
globs:
- "frontend/**/*.tsx"This makes it difficult to apply deterministic instructions to specific parts of a monorepo.
Examples:
- Frontend React rules only for
frontend/** - Python backend rules only for
backend/** - Infrastructure rules only for
terraform/** - Different testing workflows per package
Today this requires putting conditional logic inside AGENTS.md such as:
When editing files inside frontend/:
- Follow frontend conventions
When editing files inside backend/:
- Run py_compile before finishingThis works but relies on the model correctly interpreting file paths rather than the harness enforcing the scope.
Proposed Solution
Introduce a rules/ directory with optional frontmatter:
.opencode/
└── rules/
├── frontend.md
├── backend.md
└── infra.md
Example:
---
globs:
- "frontend/**/*.tsx"
- "frontend/**/*.css"
---
Use React hooks.
Prefer functional components.
Run npm test before completion.---
globs:
- "backend/**/*.py"
---
Run pytest before completion.
Use Ruff formatting.
Avoid synchronous IO.Benefits
- Deterministic instruction loading
- Reduced prompt size compared to loading all rules globally
- Better support for monorepos
- Easier migration from Cursor
- More predictable agent behavior
Alternative
If full glob support is undesirable, even a simpler mechanism such as:
{
"instructions": {
"frontend/**": "frontend-rules.md",
"backend/**": "backend-rules.md"
}
}would solve most use cases.
Additional Context
This feature would complement Skills rather than replace them:
- Skills → semantic behavior loading
- Rules → deterministic file/path-specific constraints
These two mechanisms solve different problems and can coexist.
Metadata
Metadata
Assignees
Labels
No labels
Activity
shrijayan commentedon Jul 17, 2026
One thing that would significantly improve monorepo workflows is some form of path-based instruction loading.
Semantic skill loading is useful, but file paths are an explicit signal that the harness already knows with certainty. For example:
frontend/**should load frontend conventions.backend/**should load backend rules.infra/**should load infrastructure guidance.This doesn't necessarily need to be Cursor-style
globsyntax. Instead, using a simplepathorpathskeyword would be easier for everyone to understand and more intuitive. Sinceglobis associated with Cursor, apath-based keyword from Claude Code would make the idea clearer and more accessible.Semantic matching answers the question "what am I working on?", while paths answer "where am I working?" — both are useful signals and solve different problems.
Even a lightweight solution based around a
pathorpathskeyword, or path mappings, would cover many real-world monorepo use cases.shrijayan commentedon Jul 22, 2026
@jlongster because of this change I am able not to push this inside my organization can you please helping in this.
github-actions commentedon Sep 21, 2026
To stay organized issues are automatically closed after 60 days of no activity. If the issue is still relevant please open a new one.