You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Named and multiple concurrent sc ws sessions, and workspace re-fork (P20) #215
sc ws ships one unnamed session per repo (indexed under .sc/ws/<i> with a cache file .sc/ws/cache-<i>). Named or multiple concurrent sessions, and re-forking or refreshing a workspace from a newer tip, are deferred in ROADMAP.md (P20).
Needs a design decision. Questions to settle:
Naming and addressing: sc ws fork --name <n> versus index-only; collision rules; what list, run, harvest and abandon take.
Concurrency: several sessions harvesting into one branch (harvest lands on the checked-out branch today, and falls back to work-<i> on conflict).
Refresh: rebase the workspace onto a newer tip, or abandon and re-fork. The partial-clone view rules from ADR-0050 must hold.
Interaction with the protected caches and the sparse view each workspace is forked under.
Acceptance criteria (once designed)
Two sessions can be forked, run and harvested independently with deterministic outcomes.
Tests for name collisions, harvest order, abandon of one while another runs, and a partial clone.
PR #258 builds named and concurrent sessions (ADR-0055): sc ws fork --name, --session on the other verbs, separate .sc/ws-<name>/ roots, first-harvest-wins, and ws run no longer holding the repo lock.
Remaining for this issue: workspace refresh (move a live workspace onto a newer tip). It is designed in ADR-0055 (replay each workspace's candidate with the ADR-0049 engine, refuse on any conflict, re-materialize under the recorded view) but not built: it needs a headless merge-into-tree entry point and a re-materialize that keeps files an agent may have open. Abandon and re-fork covers it for now.
Activity
tonibergholm-codento commentedon Oct 20, 2026
PR #258 builds named and concurrent sessions (ADR-0055):
sc ws fork --name,--sessionon the other verbs, separate.sc/ws-<name>/roots, first-harvest-wins, andws runno longer holding the repo lock.Remaining for this issue: workspace refresh (move a live workspace onto a newer tip). It is designed in ADR-0055 (replay each workspace's candidate with the ADR-0049 engine, refuse on any conflict, re-materialize under the recorded view) but not built: it needs a headless merge-into-tree entry point and a re-materialize that keeps files an agent may have open. Abandon and re-fork covers it for now.
fix(repo): track the ws run child in the busy marker and create it ex…
fix(repo): remove the busy marker if its pid write fails; warn if the…
docs: renumber the named-sessions ADR to 0056 (ref #215)
feat(repo,cli): named concurrent sc ws sessions (#258)
fix(repo,cli): recheck files at write time and mark an unfinished ws …
fix(repo): refuse an identical ignored file and an unemptied empty di…
fix(repo): keep an earlier crash's pending marker and count the first…
fix(repo): stop ws refresh when a parent became a symlink after the p…
fix(repo): refuse ws run while a refresh is unfinished (#215)
fix(repo): never remove a directory that replaced a tracked file in w…
fix(repo): resume an interrupted ws refresh from recorded snapshots (#…
fix(repo): refuse a protection downgrade, scope the staging sweep, na…
fix(repo): check protection downgrade before skipping an unchanged en…
fix(repo): check the tip's own protection, resume protected files by …
docs: record two ws refresh limits (#215)
fix(repo): refuse a ws refresh whose .scignore would hide an agent's …
fix(repo): read a protected .scignore, restore the cache entry on a r…
fix(repo): open the candidate's protected blobs on a retry, save cach…
fix(repo): refuse a ws refresh that stops hiding an untracked file (#215
feat(repo,cli): sc ws refresh and crash-safe resume (#268)