Structured Context Rules (.sidecarrules)

Control which files SideCar includes in its context using a .sidecarrules file in your workspace root. Rules use glob patterns to influence file scoring during context building.

File format

Create a .sidecarrules file in your project root with a JSON object containing a rules array:

{
  "rules": [
    { "type": "prefer", "pattern": "src/core/**", "boost": 0.4 },
    { "type": "ban", "pattern": "**/*.generated.*" },
    { "type": "require", "pattern": "**/*.test.*" }
  ]
}

Rule types

prefer

Boosts the relevance score of matching files so they rank higher in context.

{ "type": "prefer", "pattern": "src/api/**", "boost": 0.3 }

• boost is optional (default: 0.3)
• Higher boost values push files further up the ranking
• Useful for prioritizing the parts of the codebase you’re actively working on

ban

Removes matching files from context entirely. They won’t be included regardless of their relevance score.

{ "type": "ban", "pattern": "**/*.generated.*" }

Use this for:

• Generated code that shouldn’t be edited directly
• Large vendored files that waste context budget
• Deprecated modules you want the agent to ignore

require

Ensures matching files have at least a minimum score so they’re not dropped during top-K selection, even if they don’t match the current query.

{ "type": "require", "pattern": "src/types/**" }

• boost is optional (default: 0.1) and sets the minimum score for files that would otherwise score 0
• Does not affect files that already have a positive score

Glob pattern syntax

Patterns support standard glob syntax:

Examples

Focus on a specific module

{
  "rules": [
    { "type": "prefer", "pattern": "src/payments/**", "boost": 0.5 },
    { "type": "ban", "pattern": "src/legacy/**" }
  ]
}

Exclude build artifacts and always include types

{
  "rules": [
    { "type": "ban", "pattern": "dist/**" },
    { "type": "ban", "pattern": "coverage/**" },
    { "type": "require", "pattern": "src/types/**" }
  ]
}

How rules interact with other scoring

Context rules are applied after heuristic and semantic scoring:

1. Heuristic scoring — file extension, path matching, conversation history
2. Semantic scoring — ONNX embedding cosine similarity (if enabled)
3. Context rules — prefer/ban/require applied to the blended scores
4. Top-K selection — highest-scoring files selected within the context budget