Skip to main content
dupcanon search is retrieval-only. It does not write to the database, create candidate sets, or trigger close actions.

​
What it does

  • Runs a one-shot semantic search over a repo’s issues/PRs.
  • Supports query-driven or anchor-driven retrieval (--query or --similar-to).
  • Adds optional include/exclude semantic constraints with configurable behavior.
  • Defaults to intent embeddings and falls back to raw embeddings when intent data is missing.

​
Quick start

# simple natural-language query
uv run dupcanon search --repo openclaw/openclaw --query "oauth error" --type issue

# anchor-driven search
uv run dupcanon search --repo openclaw/openclaw --similar-to 1234 --type issue

​
Include/exclude constraints

Use --include and --exclude to bias or filter the results. 
uv run dupcanon search \
  --repo openclaw/openclaw \
  --query "websocket failure" \
  --include "auth" \
  --exclude "android" \
  --include-mode boost \
  --include-weight 0.15
Include modes:
  • boost (default): soft rerank by score + include_weight * include_score
  • filter: hard gate requiring include_score >= include_threshold
Threshold defaults:
  • --min-score: 0.30
  • --include-threshold: 0.20
  • --exclude-threshold: 0.20

​
Debugging constraints

Enable per-hit constraint diagnostics: 
uv run dupcanon search \
  --repo openclaw/openclaw \
  --query "oauth error" \
  --include "token" \
  --debug-constraints
When enabled, results show IncMax / ExcMax scores and the JSON output includes constraint_debug.

​
Output formats

  • Default: Rich table with ranks, scores, and URLs
  • --json: full JSON payload to stdout
  • --show-body-snippet: include a truncated body snippet in the table

​
Source selection

  • Default source is intent (--source intent).
  • If intent embeddings are missing, the command falls back to raw embeddings and reports source_fallback_reason in the output.
  • To force raw mode: --source raw.