Workspace recover: revive Tower-managed architects after machine reboot #830

On Sat, May 23, 2026 at 4:12 PM Amr Elsayed ***@***.***> wrote: *amrmelsayed* created an issue (cluesmith/codev#830) <#830> Context Follow-up to #829 <#829>, which built afx workspace recover for builders. The recovery flow (process revival + conversation-context restoration via claude --resume <uuid>) is now in place but only covers builders. Tower-managed architects die in the same machine-reboot scenario and have no equivalent recovery path. Problem Two architect spawn surfaces exist today: - *afx architect* — runs Claude directly in the user's current terminal. No Tower dependency. If the machine reboots, the user just opens a new terminal and runs afx architect again. *Not in scope* for recovery. - *afx workspace add-architect* (Spec 755) and the *main architect* spawned implicitly by afx workspace start — Tower-managed PTYs, persisted in terminal_sessions, killed by reboot. *In scope.* After a reboot of a workspace with several Tower-managed architects (main + named siblings per Spec 755), the user has to manually re-spawn each one. There's no equivalent of afx workspace recover for them, and the existing recovery command excludes architects by predicate (type === 'builder' filter on session rows, and porch state is the entry point — architects don't have porch projects). Proposed mechanism Reuse the on-disk Claude conversation discovery from #829 <#829>: each architect's prior conversation lives at ~/.claude/projects/<encoded-workspace-cwd>/<uuid>.jsonl. After reboot, findLatestSessionId(workspacePath) returns the most recent UUID; we relaunch the architect via claude --resume <uuid> to restore the conversation in addition to the process. Two architect kinds, both Tower-managed: 1. *Main architect* — one per active workspace, spawned by afx workspace start. Its cwd is the workspace root. 2. *Named sibling architects* (Spec 755) — additional architects added via afx workspace add-architect [--name <name>]. Each is tracked by a name; the spec docs in codev/specs/755-multi-architect.md (or wherever) detail naming rules. Their cwd may also be the workspace root, or may be a per-architect subdirectory — needs verification. Sketch of a new command (or extension of the existing one): afx workspace recover-architects [--apply] [-y] Or, more unified: afx workspace recover [--builders] [--architects] [--apply] [-y] Either reuses findLatestSessionId() from agent-farm/utils/claude-session-discovery.ts. Open design questions 1. *Unified command or separate?* Extend afx workspace recover to handle both (with optional filters), or ship a sibling recover-architects? Unified is fewer commands; separate keeps each predicate simple. 2. *Predicate for architects.* Builders use porch state for eligibility (non-terminal phase, etc.). Architects have no porch project. Likely the predicate is just: terminal_sessions row exists with type = 'architect' AND shellper PID is dead. Simpler than the builder predicate, but worth thinking through stale-architect cases. 3. *Main architect's cwd resolution.* Confirm that the main architect's working dir on respawn is the workspace root (so the jsonl lookup at ~/.claude/projects/<encoded-workspace-root>/ works). If the architect runs in a per-name subdirectory, the encoding changes. 4. *Spec-755 naming on revival.* When re-spawning a named architect ( architect-2, reviewer-bob, etc.), need to preserve the name so the user's references still resolve. The name comes from the terminal_sessions.label or a related field — verify the column. 5. *Architect spawn API.* Tower-managed architects are spawned server-side (workspace-add-architect calls client.addArchitect() rather than going through spawn.ts). The revival path needs an equivalent server call, or a flag like --resume-session <uuid> on the existing addArchitect endpoint. 6. *Concurrent revival with builder recovery.* If both are running, ensure neither steps on the other (separate terminal_sessions rows, no shared mutations). Acceptance criteria - After reboot, running the chosen command revives every Tower-managed architect that was previously alive, with conversation context restored (each lands in the same Claude session it was in before). - Standalone afx architect (local, non-Tower) remains unchanged. - Named architects keep their names on revival. - Dry-run by default; --apply actually respawns; -y skips confirmation — same UX as builder recovery. - Tests cover: main architect alive vs dead, named sibling alive vs dead, missing jsonl (fallback to fresh session), Tower not running, no architects to recover. References - #829 <#829> — builder recovery + the findLatestSessionId helper at packages/codev/src/agent-farm/utils/claude-session-discovery.ts - Spec 755 — multi-architect support (named siblings) - packages/codev/src/agent-farm/commands/workspace-add-architect.ts — current architect spawn surface - packages/codev/src/agent-farm/commands/architect.ts — local (non-Tower) architect, out of scope — Reply to this email directly, view it on GitHub <#830>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAUXGUF3G4DSUKW6E2RQOUD44IV6VAVCNFSM6AAAAACZKWLNIKVHI2DSMVQWIX3LMV43ASLTON2WKOZUGUYDSOBWGUYDAMA> . Triage notifications on the go with GitHub Mobile for iOS <https://apps.apple.com/app/apple-store/id1477376905?ct=notification-email&mt=8&pt=524675> or Android <https://play.google.com/store/apps/details?id=com.github.android&referrer=utm_campaign%3Dnotification-email%26utm_medium%3Demail%26utm_source%3Dgithub>. You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>