Disk hygiene: run logs grow unbounded and campaign worktrees are never reaped

[roninjin10]

A workspace hit ENOSPC (926GB volume 100% full) during a live mission. Post-mortem found two smithers-owned growth vectors:

1. Run logs are unbounded: .smithers/workflows/*.log accumulated 998 files, many 100MB+, with no rotation, size cap, or retention policy. Detached runs (up -d, oneshot) append agent stream output indefinitely.
2. Workflow worktrees are never reaped: .smithers/workflows/.worktrees/ held 51 worktrees (~190G with node_modules) from a campaign whose PRs had all been pushed a day earlier. Nothing prunes worktrees when a run finishes, and smithers worktree doesn't cover this hidden directory.

Proposals:

• Log rotation/retention (e.g. cap per-run log at N MB with truncation notice; smithers gc or automatic reap of logs for finished runs older than N days)
• Worktree reaping on run completion (opt-out for debugging), and smithers worktree list/reclaim should surface .smithers/workflows/.worktrees
• A smithers doctor/gc disk report so the workspace can warn before ENOSPC kills live runs (ours orphaned a 44-minute mission run)

🤖 Generated with Claude Code

https://claude.ai/code/session_01L7UGBH6syDZeMmiHvPJGk5

[roninjin10]

Two more unbounded growth vectors confirmed (3rd ENOSPC incident in 24h on the same workspace):

3. Lane/sandbox scratch under /private/tmp/smithers/: 63G accumulated in ~2 days of mission lanes (worktree isolation dirs, npm installs, playwright artifacts). Nothing reaps finished lanes' scratch.
4. upgrade-smthrs*.previous* dirs in /private/tmp: ~46G of retained previous-version copies from a single upgrade-dependents campaign, never cleaned after the campaign closed.

Suggested smithers gc scope: finished-run sandboxes + campaign .previous dirs + rotated logs + reaped worktrees, with an age/live-run guard.

Related external note: OpenCode's session DB (~/.local/share/opencode/opencode.db) grew 24G→29G in one day under heavy kimi lane usage — separate upstream issue, but it compounds the same disk pressure.