ontograph/ontoindex/ontoindex-navigation

OntoIndex Navigation

Always query OntoIndex first using the task-appropriate tool below. Repo queries schedule one managed background refresh when HEAD differs from the indexed commit, or when a repository without commits has changed files in its recorded indexed scope. The response uses the existing index with a stale/refreshing warning. Use stale/refreshing results, label claims with that freshness, and verify edited or new code from source. Uncommitted edits use the automatic overlay; check coverage warnings and inspect uncovered source directly. Never fall back to grep-only just because the index is stale. Automatic refresh keeps the recorded scope, excludes huge-repo-symbols indexes, and is rate-limited by ONTOINDEX_AUTO_REFRESH_INTERVAL_MS (default 300000 ms). ONTOINDEX_AUTO_REFRESH=0 disables it. Only the coordinating session may run the installed ontoindex analyze manually when immediate refresh is required and no refresh job is active; check the returned refresh/job status first. Sub-agents never start reindex jobs; refresh-on-query covers them.

Tool risk output (LOW/MEDIUM/HIGH/CRITICAL) is authoritative; warn on HIGH/CRITICAL before editing. With impact, maxDepth: 1 returns direct dependents to review first; membership alone does not prove breakage. If tools are missing from the session, search the tool catalog for ontoindex. If tools remain unavailable, a call fails, or a result reports degraded/untrusted evidence, say exactly: OntoIndex stale/degraded: <reason>; source inspection is authority. Continue with usable OntoIndex evidence and bounded source inspection; use bounded rg and file reads when tools cannot answer. A stale warning alone does not trigger this failure path. Never present rg or git diff as OntoIndex evidence.