[kixelated]

Problem

The incident showed mbx collecting one “live” target and draining shared objects below gc.max_size when free space fell below its default floor. “Live” means that the checkout still exists, not that a command holds its target's lease. This PR completes quest/m0/ci/mbx-gc.md and removes its CI epic link.

Approach

Source configuration: kixelated/dotfiles#36, commit 26d375495bfad24cdac99f035cc67e35d7e9a045. Land that source PR before this bookkeeping PR. The policy is already active through the existing dotfiles config symlink on desktop.

• target.keep = ["/"] retains targets for every existing checkout, including direct binary execution and gaps between launches. Abandoned targets remain collectible.
• gc.min_free_size = "none" disables pressure-driven eviction; normal cache budgets remain enabled. A build can fail with the ordinary No space left on device I/O error instead of deleting targets or draining the shared cache to accommodate unrelated disk use.
• Dotfiles documents inspection and ownership-aware cleanup. The live checkout's unrelated changes and commits are preserved; its local source checkpoint is 75d60f6b495505329bf8babc77356fb70b6e9dbb. Its publication branch contains only this configuration and documentation.

Why the protection was absent

mbx 1.25.0 protects a running wrapped command with a shared view lease, and native Cargo with Cargo's build lock. Directly launched binaries and intervals between launches hold neither. The current managed-target documentation and upstream process-protection change confirm that even mbx 1.29.2 deliberately ignores directly launched programs under low-space collection. The incident's aggregate log does not name the victim checkout, so its exact command/lease cannot be reconstructed; it is not evidence that a held lease failed. The isolated reproduction establishes the unprotected path and the configuration protects it without needing that missing attribution. This is supported policy, not an unresolved upstream defect; no upstream post is needed.

Disk attribution, desktop, 2026-11-03

Read-only snapshot while other work continued; rounded allocated-byte readings can change during the scan and shared blocks are not additive:

• Root filesystem: about 1.8 TiB used, 59 GiB initially available (97%).
• MoQ worktrees: about 516 GiB. The three largest sampled .tmp/ trees held 15, 14 and 13 GiB; these are scratch, separate from centrally managed target views. No ownership was inferred from size.
• mbx cache: about 253 GiB, including 108 GiB actions, 103 GiB targets, 41 GiB incremental state and 1.2 GiB scheduler data.
• Other large consumers: about 78 GiB uv cache, 174 GiB .local, and 248 GiB under ~/work.
• Nix store: du counted 173 GiB allocated but could not read an active Nix chroot, so this is a lower bound. Its database lists 99,703 registered paths and 486.9 GiB NAR size; NAR size is serialized logical size, not disk allocation.
• quest worktree status has no registered checkouts on this host. Automatic quest cleanup therefore cannot be assumed to reclaim these T3 worktrees. No other worktree, shared cache or Nix path was deleted.

Impact

• Public API and wire: none. No MoQ runtime or harness change.
• The source policy keeps all existing targets, including idle ones, until checkout retirement or explicit cleanup. They may exceed the target budget. It preserves outputs over reclaiming headroom; there is no early free-space refusal or reserved capacity. Ordinary build I/O failure is the chosen failure policy.
• Other agent hosts receive the policy when their dotfiles are updated. No deployment, merge or PR monitoring is part of this attempt.

Alternatives and choices

Lowering the threshold merely postpones the same deletion. The latest upstream process scan still excludes low-space collection and has platform/snapshot limits. Holding Cargo open would miss directly launched programs and intervals between launches. Retaining existing checkouts and disabling pressure eviction requires two supported settings, without a new launcher, threshold or dependency.

Validation

• Installed mbx 1.25.0 and SHA-256-verified latest stable 1.29.2: private XDG_CONFIG_HOME, MBX_CACHE_DIR and MBX_TARGET_ROOT throughout. A copied executable communicates over a pipe to prove it is running; fixture records use an old timestamp instead of waiting. Explicit GC never touched host state.
• Both versions: unprotected running target deleted under forced gc.min_free_size = "100TiB"; shared-lease target preserved; tracked-config target preserved under forced pressure and separately a 1B target budget; abandoned target collected in all cases.
• Real tiny Cargo builds through each version succeed. A subsequent build-script write to /dev/full fails with exit 101 and explicit No space left on device, retaining the earlier executable. GC with the tracked policy preserves the populated action store and live target.
• Dotfiles publication snapshot and unchanged main each run 71 tests with 70 passing and the same pre-existing cleanup shell-dispatch missing-log error. Live checkout: 88/89, same error. No full-suite pass is claimed.
• quest gate nix develop --command bash -c 'export PATH=<pinned Nix Rust>/bin:$PATH; exec just ci origin/main' passed in 94 seconds: all selected lint/tooling lanes, repository tooling tests and 600-document quest validation. git diff --check origin/main...HEAD passed. No MoQ runtime package changed, so no interop run is needed.

Review

Skipped for 4b5c0bdbe9cd6d568217948e4a68a516d8557530: this committed MoQ diff deletes the completed quest and its single epic link. The two supported dotfiles settings and their guide have the separately recorded manual source audit and released-binary behavior checks. No sub-agents were started.

Follow-ups

• Investigate the pre-existing dotfiles cleanup test's missing stub log. No matching open fix was found.
• Reclaim finished T3 worktree scratch through the existing cleanup workflow after verifying ownership and preserving local/submodule state. The empty quest-worktree registry means it is not safe to assume these checkouts are registered. This task does not delete active targets or introduce a new disk budget.
• No new quests. moq-dev/moq#7715 remains the independent interop workaround; this PR does not absorb it or any ingest7808 work.

(Written by GPT-6.1 Sol)

[coderabbitai]

Warning

Review limit reached

You've used all free OSS reviews for now. Wait for the free limit to reset to keep reviewing this public repository.

Next included review available in 26 minutes.

Check out review usage here.

View limit details

Limit details: You’ve used all 4 included reviews currently available.

Learn how review limits work.

Review configuration:

Run configuration

• Source mode: Auto
• Repository: Unavailable
• Branch: Unavailable
• File: No configuration file from this repository applied
• Commit: Unavailable
• Layers: Organization UI
• Review profile: CHILL
• Plan: Advanced
• Run ID: 22c9c77f-234b-4b50-9463-ebd35bea4973

Commits

Reviewing files that changed from the base of the PR and between 4506e3f and 4b5c0bd.

Files selected for processing: 2

• quest/m0/ci/README.md
• quest/m0/ci/mbx-gc.md

• Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

Share

• X
• Mastodon
• Reddit
• LinkedIn

[pullfrog]

✅ No new issues found.

Reviewed changes

• Completed quest removed — quest/m0/ci/mbx-gc.md is deleted, consistent with how prior completed quests have been retired in this repo.
• Epic index updated — its single entry is removed from quest/m0/ci/README.md.

No references to mbx-gc or to the deleted file's Related link remain in the repo, and the README edit is a clean single-line removal. The policy change itself lives in kixelated/dotfiles#36, so this checkout has no runtime, API, or wire surface to review.

  ^{｜ View workflow run ｜ Using deepseek-v4.1-flash-1017 (free via Pullfrog for OSS) ｜ 𝕏}

[kixelated]

Automated review by review (OpenAI)

Reviewed head 4b5c0bd. I found no additional actionable issue in retiring the quest. As the PR notes, the source change in kixelated/dotfiles#36 must land first; it is still open at this review.

Verification: GitHub-only static review of the quest, linked configuration and upstream retention behavior. I did not run the reported tests or verify the desktop deployment.

(Written by review (OpenAI))