Use when investigating failed, stuck, cancelled, or suspicious Orbit job runs. Guides agents through run state, step summaries, v2 audit events, logs, blobs, process state, task linkage, recovery/follow-up, and safe reporting.

Use this when executing an existing Orbit task or carrying a human request through the Orbit task lifecycle with explicit status tracking.

Use when navigating and inspecting codebase via the knowledge graph instead of raw file reads.

Use this when creating, searching, updating, superseding, pruning, or auditing Orbit project learnings. Agent tools cover `orbit.learning.add/show/update/upvote/supersede/comment.add/comment.list` plus `orbit.search`; list/sync/prune/comment.delete are CLI-only operator workflows. Covers scope-OR matching (path globs / tags), evidence shape, the `update` vs `supersede` boundary, the YAML-source-of-truth + SQLite-index model, and why never to edit `.orbit/learnings/` files directly.

Use when searching tasks, docs, learnings, or ADRs through the unified `orbit search` query surface; also covers the secondary `orbit semantic` lifecycle namespace for installing, removing, inspecting, and indexing the local embedding companion.