afx cleanup can't reach orphaned Tower terminals after porch done self-completion (registry-keyed cleanup, no fallback)

[amrmelsayed]

Problem

When a builder runs porch done <id> to mark its protocol complete (BUGFIX, AIR, PIR — any protocol that ends with a porch done rather than awaiting an external signal), the builder is deregistered from afx's project/issue registry as part of completion. Subsequent afx cleanup attempts then fail because cleanup is registry-keyed — both afx cleanup -p <id> and afx cleanup -i <issue> return:

[error] Builder not found for project: <id>
[error] Bugfix builder not found for issue #<issue>

The Tower-level PTY session for the builder is not killed by porch done (Tower's terminal registry is independent of the project/issue registry). It remains visible in afx status as an (active) terminal indefinitely. There is no registry-independent terminal-kill command in afx.

Observed

Sequence on a real BUGFIX builder run that completed cleanly:

1. Builder ran porch done <id> (BUGFIX protocol-complete on merged PR).
2. Builder sent the "ready for cleanup" notification.
3. afx cleanup -p <id> → Builder not found for project: <id> (builder already deregistered).
4. Manually git worktree remove --force + git branch -D builder/<...> succeeded (those don't go through afx's registry).
5. afx cleanup -i <issue> -f → Bugfix builder not found for issue #<issue> (same; cleanup still can't find a registry entry).
6. afx status continued to show the builder's terminal as (active) with no work, no worktree, no branch, no porch state — a dead PTY in Tower with a stale registry entry.

The only afx mechanism that can clear an orphaned Tower terminal is afx tower stop && afx tower start — which kills every other live builder terminal in the session. Unacceptable collateral in any multi-builder workflow.

Impact

• Every self-completing builder accumulates a permanent orphan terminal. afx status becomes increasingly noisy with dead entries from past completed work, with no way to distinguish dead from live except by external knowledge.
• No clean recovery exists. Killing the underlying PTY process directly bypasses Tower and risks Tower-state inconsistency. Restarting Tower wipes all live work. The user is left with permanent cosmetic noise that compounds across sessions.
• Cleanup ordering is fundamentally lossy. A builder that self-completes via porch done is structurally uncleanable via afx the moment the registry entry is dropped — there is no window in which afx cleanup can both find the builder and kill its terminal.

Related

• Loose family: porch: verify-wedge — a reworked artifact can't re-enter verify after its approval gate has fired (rollback can't target current phase) #776 (porch verify-wedge) — both are afx/porch lifecycle gaps where a state transition leaves an unreachable artifact.

[amrmelsayed]

Sibling: #801 (porch approve state-push silently fails after auto-delete on PR merge — builder worktree never sees gate-approved state-sync). Same lifecycle family as this issue — both leave the builder in a stale / unreachable state after a normal completion path.

[amrmelsayed]

Sibling: #802 (state-bookkeeping commits accumulate as orphan artifacts on builder branches when not auto-deleted on PR merge). Same afx/porch-lifecycle family — branch persists with dead bookkeeping commits after work is done.