PR Duplicate Command
Detect duplicate pull requests using semantic search across your issues and PR collections.Syntax
Options
| Option | Short | Type | Description | Default |
|---|---|---|---|---|
--repo | string | Repository (owner/name) | Required | |
--number | -n | number | PR number to check | Required |
--top-k | number | Maximum candidates to return | 5 | |
--threshold | number | Minimum similarity score (0.0-1.0) | 0.65 | |
--token | string | GitHub token (falls back to GITHUB_TOKEN) | - | |
--config | -c | file | Path to configuration file | .github/simili.yaml |
--dry-run | bool | Skip Qdrant search (returns empty candidates) | false | |
--help | -h | bool | Show help message | - |
Examples
Check PR for duplicates
Adjust similarity threshold
Return more candidates
Dry-run mode
How it works
- Fetches PR details and changed file paths from GitHub
- Embeds the PR content:
Title: ...\n\nBody: ...\n\nChanged Files:\n- path/a - Searches both the
collection(issues) andpr_collection(PRs) in Qdrant - Deduplicates and sorts candidates by similarity score
- Optionally runs an LLM-based duplicate verdict on the top-3 issue candidates
- Returns a JSON result with candidates and duplicate assessment
Output
Configuration
To enable a dedicated PR collection, setqdrant.pr_collection in your config:
pr_collection is not set, PRs are stored alongside issues in the main collection.
Indexing PRs first
Before runningpr-duplicate, make sure PRs are indexed:

