Auwai

Workspace

A workspace is a virtual repository. It is the root of the same partition tree that a single repository uses.

Shape

workspace
  repo A
    partition
      segment
  repo B
    partition
      segment

A workspace keeps each repository index separate. It adds a target, findings, decisions, and a handoff brief. The survey combines the first-party chunks into one tree with the repository name as the first path segment.

Survey

auwai workspace survey --workspace NAME --context-window N prints the tree and the segment plan. --json includes the tree and the plan.

The plan uses the same segment planner as a single repository. A read segment can span repositories. A chunk id in the plan is qualified with the repository name.

Full-read flow

The workspace plan is separate from every repository plan. The qualified chunk id is the join key between the virtual plan and the per-repository indexes.

Full read

auwai workspace fullread plan persists the virtual plan in .auwai/workspace/fullread.json and creates a workspace coverage ledger in .auwai/workspace/coverage.sqlite. The ledger records one row per qualified chunk and one row per segment.

next, read, done, status, and verify drive the loop. A read segment cannot close until every planned chunk is complete in the workspace ledger. A synthesis segment has no source bodies and closes after the findings review. The per-repository indexes remain independent; the workspace ledger is the coverage record for the virtual repository.

Existing commands

The workspace still provides:

Current limits

Cross-repo categories are tags on the tree, not separate first-class nodes. The workspace target has no per-target coverage ledger; workspace coverage is stored in .auwai/workspace/coverage.sqlite. The full-read runner does not call a model.