repo-discovery

When this applies

• At the start of a question: confirm which repos are indexed and plan where to look before issuing any content queries.
• Mid-investigation: when the trail leads into a dependency, upstream library, or sibling repo - expand scope before drawing conclusions.
• When a dependency boundary is reached: if the current repo's code delegates to an external dependency (e.g., an import resolves outside the repo, behavior is implemented in an upstream library), always check whether that dependency is indexed before concluding the answer can't be found. Never assume a dependency is unindexed without checking.

Tool

Use sourcerer.refs.list with wildcard git_repo and git_org patterns. Each result row is one indexed ref; the distinct git.repo values in the results are the repos available to query.

Progressive disclosure - narrow first, expand only as needed

Start with the most specific pattern likely to match the question. Broaden only if the result is empty or you need more coverage.

Avoid git_repo: * without an org filter unless you genuinely need all orgs - it returns every indexed ref and consumes many tokens.

Interpreting results

From the result rows:

• Distinct git.repo values - the repos available to query.
• git.ref_type - whether each repo has branches, tags, or both. Useful for knowing whether ref-resolution will find tags or needs to fall back to a branch.
• status - a ref with status other than ready may have incomplete content.

Example: expanding scope mid-conversation

A question about Elasticsearch's query parsing leads into how Lucene's QueryParser works underneath:

1. Initial discovery: refs.list(git_org: elastic, git_repo: elasticsearch) - confirm the repo and plan the search.
2. Answer the Elasticsearch side. When the trail leads to Lucene, expand:
3. refs.list(git_org: apache, git_repo: lucene*) - discover which Lucene repos are indexed.
4. Use ref-resolution on the relevant Lucene repo to pin a commit, then query content.

Each discovery call should be targeted - retrieve only what you need to plan the next step.