mu 3.13.3

Test upkeep leaves a record. An upkeep pass now ends by committing a
dated report to the repo, so its findings outlive the workstream.

Upgrade with npm i -g @mu-crew/mu@3.13.3. New agent sessions load the updated
skill. No CLI, API, schema or extension change.

Changed

• Skill: test upkeep ends with a committed report. The
  test-health loop gains a
  sixth step: a dated docs/test-health/ report in the repo, holding the
  findings and every refuted or rejected verdict, so the next pass starts
  from them instead of from scratch when the workstream's notes are gone.

mu 3.13.2

Keep verification useful as projects grow. The mu skill now defines
fast/slow gate responsibilities, scoped evidence reuse and routine test upkeep.

Upgrade with npm i -g @mu-crew/mu@3.13.2. New agent sessions load the updated
skill. Existing workers need updated briefs; no schema or extension change.

Changed

• Skill: fast/slow gates and routine test upkeep. Canonical
  test-health guidance assigns targeted
  worker/reviewer checks, equivalent gate records and one full merged-tree
  gate per batch. Milestones and budget crossings prioritise upkeep without
  a crew-wide barrier; auth/build isolation and required pre-push checks stay.
  mu's own gate commands remain unchanged pending review.
  Pilot baseline supplied for swaywardlock 90c3f1c: cold full 557.885 s,
  fast 216.140 s, slow 341.696 s (e2e 255.561 s, release 82.414 s);
  223 unit tests ~1 s, 64 e2e cases / 422 assertions, no skips. Converters
  initially cost ~28 s in fast: fast is not universally seconds. This is
  pilot timing, not an equivalent gate record for mu. A separate targeted
  vibrancy-wait follow-up preserved all seven assertions and reduced that
  case's mean from 3.350 s to 2.140 s; it does not establish a full-gate
  speedup or a solved budget.

mu 3.13.1

jj 0.46: recreating a deleted workspace works again. A spawn that reuses
the name of a jj workspace whose directory was deleted failed on jj 0.46.

Upgrade with npm i -g @mu-crew/mu@3.13.1. No schema or extension change.

Fixed

• A jj workspace whose directory was deleted can be recreated (jj 0.46).
  mu agent spawn --workspace forgets a stale jj workspace of the same
  name first. jj 0.46 keeps printing the deleted path as the workspace
  root, so mu missed it, and in a colocated repo the workspace is also a
  Git worktree that jj workspace forget leaves registered. mu now checks
  the root on disk and runs git worktree prune after the forget, so the
  re-add no longer fails with "Workspace named … already exists" or
  "missing but already registered worktree".

mu 3.13.0

Faster everyday verbs. tmux-touching verbs no longer load execa (about
50-60 ms each), pure-JS dependencies are bundled, and a bare verb in a
mu-spawned pane skips the tmux workstream lookup. mu doctor counts ops by
key range (~250 ms less), peer sync resumes from where it stopped, and the
TUI launches faster and stops re-detecting the VCS every 10 s.

Upgrade with npm i -g @mu-crew/mu@3.13.0, then /reload in running pi
sessions to pick up the extension. No schema change. Task, track and peer
lists now sort names by code unit (a-b, a0, a_b; uppercase first).

Changed

• Spawn warns on Claude's folder-trust dialog. The dialog defaults to
  "No, exit", so the Enter that ends --send quit claude. The tmux
  startup scan now spots it (non-fatal): the pane and agent stay, the
  initial --send is skipped (send.error TrustDialogPending in
  --json), and a warning plus next step says to answer it in the pane
  and then mu agent send. It does not guard a later agent send or the
  herdr spawn path, and mu never presses a key for you.

• TUI dashboard fills its space. A card's maxRows now caps only
  the first pass of the row allocator: leftover rows go to cards whose
  data still overflows, in proportion to the rows each hides, so a
  40-item Ready list at 80 rows shows all 40 instead of 15 above 60
  blank rows. Cards also move between columns: arrangeColumns
  (replacing balanceColumns, which only ran after a cull) searches
  single moves and swaps for the fewest hidden rows, even with all ten
  cards visible, and keeps the previous layout unless the new one hides
  max(3, 10%) fewer rows. The search also descends from the previous
  layout, so a retained layout never has a single move or swap that
  beats it by more than that margin. Among layouts that hide the same rows, the search
  prefers fewer cards away from their default column, then slot order. Footers and mouse hit regions read the same
  budgets (columnRowBudgets).

• Skill: recipe wording from a prior runtime. An unknown owner
  is not a dead one: orchestrator-loop says to read the pane or run
  mu doctor before releasing its task. Reviews with only
  UNVERIFIED items go to the human, not into a REJECT round. findings § Record
  grades severity strictly (critical needs a concrete scenario).
  ultrathink and loop-until-done refute the umbrella's criteria on the
  merged tree before close --if-ready.
  watcher caps attempts per item and parks it with every attempt listed.
  long-run records DEAD-END: notes on the umbrella.
  backlog-triage re-verifies an item against the code before starting.
  dag.md states that every edge comes from a verb the caller ran (block, add/reparent --blocked-by) and no write adds edges on its own.
  VISION cites a prior runtime's transport and sentinel-marker lessons.

• A changed peer segment is ingested from where the last read stopped.
  Every append used to make each other machine hash the whole segment,
  read it a second time as text and split it (~120 ms for this fleet's
  64 MB segment, ~9 ms of it needed). A clean read now saves the byte
  offset and running sha256 state next to the caught-up fingerprint; the
  next ambient ingest reads only the appended bytes, and accepts them
  only when their hash is the manifest's sha256 (~20 ms). A shrunk
  file, a stale manifest, a defect (which saves no offset) and mu sync
  take the full read and whole-file verify as before. The resumed hash
  trusts the prefix it already read: a local edit to those earlier bytes
  that keeps the file growing is caught only by mu sync, not by the
  ambient ingest. No
  schema change: the offset is extra fields in the existing
  sync_fingerprints value, which an older mu never matches and
  ignores.

• mu workstream list reads mux liveness from one list-sessions.
  It also ran has-session once per workstream (6 spawns for 6
  workstreams). The listing now reuses the session names it already
  fetched, and falls back to per-workstream has-session when
  list-sessions fails, so a transient failure still never marks live
  sessions dead. ~22 ms CLI CPU saved; output is identical.

• mu_delegate lists agents at most every 30 s while a failed delegate
  lingers. Each spawn ran mu agent list (~150 ms of CPU) only to prune
  the footer's failed count; the refresh is now throttled to once per
  30 s. Spawning and safety checks are unchanged; the footer may show a
  closed failure for up to 30 s longer.

• mu agent spawn sets the tmux pane border in one tmux call. The four
  set-option -w calls are ;-chained into one invocation and target the
  pane directly, which drops the display-message #{window_id} lookup
  (the border helper's 5 tmux processes become 1, ~33-80 ms per spawn). Options, -w scope,
  MU_BANNER_QUIET, and the swallow-all-errors behaviour are unchanged.

• The TUI detects the VCS backend once, not every 10 s. The slow tick
  ran jj root, sl root (twice) and git rev-parse before every
  git log just to learn the same backend again. It now caches the
  backend per project root for the TUI's lifetime (~4 of 5 processes per
  tick; ticks after the first dropped from ~690 to ~175 ms best case on
  a loaded machine). A none result is not cached, nor is a detection
  during which a root probe failed (spawn error, signal) rather than
  cleanly answering "not this VCS": a transient jj root failure in a
  colocated jj/git repo no longer pins git. mu state is unchanged (it
  never probes the VCS).

• The TUI bundles ink and react into its lazy chunk. Importing the TUI
  resolved and compiled ~560 ESM files from node_modules per launch. tsup
  now inlines ink, react and their pure-JS deps into dist/tui-*.js (loaded
  only by bare mu and mu state --tui; other verbs are unchanged), using
  the createRequire banner from the commander/cli-table3 bundling. NODE_ENV
  is not defined at build time, so src/cli/tui-load.ts still picks react's
  production build at runtime and an explicit NODE_ENV still wins. TUI
  import ~470 -> ~67 ms, first Agents card ~890 -> ~440 ms (PTY, 20
  alternating launches on a copied DB). react-devtools-core stays external
  (ink loads it only under DEV=true).

• Task, track and sync-peer sorts use code-unit order, not
  localeCompare. The first localeCompare call initialised ICU
  (~4 ms on nearly every list verb). Ties by name now sort by code unit,
  so a-b, a0, a_b replaces the locale order a_b, a-b, a0; uppercase
  sorts before lowercase. Sort keys, ROI order and everything else are
  unchanged.

• mu doctor and mu db forget count a workstream's ops by key range.
  The per-workstream substr(key, …) test scanned all of ops once per
  torn-down workstream (27 scans of 61k rows, ~320 ms of SQL in
  mu doctor's size hint). It now seeks idx_ops_entity_key with a key
  range per entity (~28 ms); counts and deleted rows are identical.

• A bare verb in a mu-spawned pane skips the tmux lookup. With no
  -w and no $MU_SESSION, resolveWorkstream ran tmux display-message -p '#S' (and loaded execa) even though spawn already
  put the workstream in $MU_WORKSTREAM. It now uses $MU_WORKSTREAM
  when $MU_MANAGED_AGENT=1 is also set: bare mu task list ~133 -> ~67
  ms from an agent pane. -w and $MU_SESSION still win, and a
  $MU_WORKSTREAM without $MU_MANAGED_AGENT (a stale export in a user
  shell) is ignored. User-started orchestrator panes are unchanged.

• Spawn sees a pi agent's control socket within ~50 ms. The spawn
  handshake polled every 250 ms, so a socket pi bound just after a poll
  waited up to 250 ms to be noticed. It now stats the socket path every
  50 ms (no fork) and probes as soon as it exists: readiness overshoot
  132 -> 24 ms, ~100 ms less per pi spawn. The pane-alive check and
  scrollback capture keep their 250 ms cadence, plus one final check at
  the budget's deadline, so a pane dying in the last 250 ms keeps its
  crash scrollback.

• mu_delegate closes the pane and writes the record note in
  parallel. A finished delegate with record waited for
  mu agent close before starting mu task note, though neither needs
  the other. They now overlap: ~70 ms less before the answer arrives
  (median of 25 pairs), on clean finishes that close and record. Failure
  reporting is unchanged: a refused close and a failed note are each
  named in the follow-up.

• The keep-driving nudge reads mu task list, not mu state. At
  each armed settle the pi extension ran mu state --json only to read
  IN_PROGRESS task names, paying for the tmux pane listing and the
  orphan and workspace views it never used. It now runs
  mu task list --status IN_PROGRESS --json per dispatched workstream
  (--sort recency, newest first as state listed them), in parallel,
  deduplicated by task: ~56-69 ms less per armed settle (one
  workstream). A task still owned by a dead agent shows in the nudge until the next
  mu state or mu task wait reaps it.

• Faster tmux-touching verbs. The tmux, herdr, kick and murmur-status
  executors run through node:child_process instead of lazily importing
  execa (~110 modules, 25-40 ms). mu state, mu agent list, and bare
  verbs that resolve the workstream via tmux drop by ~50-60 ms per call
  (state ~140 -> ~85 ms). Output and exit codes are unchanged; capture
  keeps execa's 100 MiB buffer cap.
  In-flight children are killed when mu exits or gets SIGTERM, SIGINT or
  SIGHUP, as execa's cleanup did, so no tmux or murmur process outlives mu.

• Faster startup: pure-JS dependencies are bundled. Every verb
  resolved, read and compiled about 35 node_modules files (commander,
  cli-table3, string-width and their dependencies). tsup now bundles
  commander, cli-table3, string-width and picocolors into dist/
  (native better-sqlite3, execa, ink and react stay external;...

mu 3.12.0

A delegate that ends without an answer leaves its evidence. Timeout,
API error, empty finish, dead pane or cancel: the follow-up names the
delegate's pi session file and copies its last message so far to
~/.cache/mu/delegate-answers/.

Upgrade with npm i -g @mu-crew/mu@3.12.0, then /reload in running
pi sessions to pick up the extension. No schema change. The session
file is reported by the delegate's own extension, so delegates need
3.12.0 too; an older one still gets the mu agent read pointer.

Added

• Delegates that end without an answer leave evidence. A
  mu_delegate follow-up for an API error, an empty finish, a
  timeout or a dead pane, and the mu_delegate_cancel result, now
  name the delegate's pi session file and copy its last message so far
  to ~/.cache/mu/delegate-answers/<ts>_<name>.partial.md. Before, the
  only pointer was mu agent read, which stops working once the pane
  is gone. Quitting pi or /reload lists each running delegate's
  session file too.
• The control socket's status reports sessionFile, so the reply
  to mu agent spawn --send / mu agent send carries it before any
  run settles.

mu 3.11.3

Escape no longer loses delegate answers; spawn and DB-open fixes.
An Escape that drops queued delegate answers leaves a pointer to each
one. A fresh DB survives a parallel fan-out, and --next-free skips
names whose workspace dir is left on disk.

Upgrade with npm i -g @mu-crew/mu@3.11.3, then /reload in running
pi sessions to pick up the extension. No schema change. Both sides of a
delegate need 3.11.3 for the session-file pointer: an older delegate
falls back to the copy and mu agent read.

Added

• Escape no longer loses queued delegate answers. Escape clears
  pi's follow-up queue, so an answer waiting there was gone. The
  extension now copies each dropped answer to
  ~/.cache/mu/delegate-answers/ ($XDG_CACHE_HOME/mu) and writes
  one line per answer into the conversation, without starting a turn,
  naming the copy and the delegate's pi session file. The copy is
  cache: the session file keeps the original.
• mu agent wait --json reports a pi agent's sessionFile, the
  transcript of the run it waited for. The control socket's wait
  reply carries it.

Fixed

• spawn --next-free skips a name whose workspace dir is left on
  disk. A dead agent's checkout outlives its row, kept as evidence.
  --next-free checked only agent rows, so it picked that name again,
  and --workspace then failed with WorkspacePathNotEmptyError
  instead of moving on. mu_delegate with workspace: true hit this
  when a label repeated. The leftover dir is untouched; the spawn takes
  the next name.
• Parallel mu calls on a fresh DB no longer drop one. The first
  fan-out into a new DB (a new machine or MU_DB_PATH) could lose a
  process two ways. Switching the file to WAL returns SQLITE_BUSY
  without honouring busy_timeout, so a loser exited db locked; it is
  now retried for up to 5 s. A process reading the schema version while
  another created the tables could see a half-built DB and refuse it as
  v1 (SchemaTooOldError); the version is now read in one transaction.

mu 3.11.2

mu_delegate accepts model ids it used to refuse.
Upgrade with npm i -g @mu-crew/mu@3.11.2, then /reload in running
pi sessions to pick up the extension. No schema change.

Changed

• mu_delegate is forgiving about model. It used to refuse any
  value outside [A-Za-z0-9._:/@+-], so real ids such as
  modelbridge/meta-openai/gpt-6.1-sol[1m]:high were rejected, and
  that happened in 12 past sessions. It now trims whitespace, drops
  surrounding quotes or backticks and a pasted --model / --model=
  prefix, and shell-quotes any other value inside --command, so pi
  decides whether the id exists. An empty model counts as unset. The
  only value still refused is one with whitespace inside it.

mu 3.11.1

pi's footer shows delegate results that have not arrived yet.
Upgrade with npm i -g @mu-crew/mu@3.11.1, then /reload in running
pi sessions to pick up the extension. No schema change.

Added

• mu_delegate footer counts results pending. A delegate that
  finished while pi was mid-turn has its answer queued as a follow-up,
  and the footer used to drop it from the count, so it looked as if
  nothing was outstanding. The footer now reads 1 result pending until
  the answer enters the conversation. Escape, which clears pi's
  follow-up queue, also clears the count.

mu 3.11.0

The skill learns model tiers, and a review panel's fixes land. A new
recipes/models.md says which model tier each role needs; a five-angle
review of the skill found 41 problems, and the confirmed ones are fixed.
mu agent close no longer deletes edits in a cp -a workspace, and
mu agent send to a dead pane says so.

Upgrade with npm i -g @mu-crew/mu@3.11.0, then /reload in running pi
sessions to pick up the skill. No schema change.

Changed

• Skill: one agent per ready task, not per track. An umbrella puts
  every unit on one track, so "one agent per ready track" read literally
  ran a fan-out with one worker. The skill now says to spawn per ready
  task, and to chain tasks that share files with block so only one is
  ready at a time.
• Skill: wording fixes. REFUTED is the verdict that sends a brief
  back for rewrite (there was no AMEND); SKILL.md drops text that
  restates --help and the outside-pi spawn steps, uses VOCABULARY
  terms, and models, worker and brief recipes get checkable Done whens.
• Skill: model tiers. New skills/mu/recipes/models.md ranks models
  by position in a lab's lineup (frontier, strong, mid, cheap, local),
  says which roles need which tier, and keeps local models as a last
  resort. Checkers run at the tier of whoever wrote what they check, or
  higher, from another family. SKILL.md and ten recipes point at it in
  place of the pi_mini/pi_big keys. Skill-only: /reload in running
  pi sessions to pick it up.

Fixed

• Skill: review recipes keep blockers honest. UNVERIFIED findings
  no longer contradict --if-ready (umbrella stays open);
  adversarial-review re-blocks downstream work on review_x_2 after a
  rejection; findings Group blocks the umbrella on the fix before
  superseding; adversarial-review uses finding and scopes itself to
  commit-producing units; rules-audit splits delegate/workstream mode;
  tasks-or-calls defines delegate call for scratch and a satisfiable
  Done when.
• Skill: drift-audit and recovery. The drift-audit late-note query
  no longer hides an unrelated CLOSE: note that follows a close;
  recovery's Done when is per operation and reachable on the DRIFT path;
  its teardown/undo bullets keep only what --help lacks.
• Skill: codemode driver. The scratch script writes each note to a
  file and pipes it to mu task note, so answer text (e.g. a MU_EOF
  line) never runs as shell; the stale cap-refusal warning points to
  orchestrator-loop § Concurrency.
• Skill: remote-workers recipe safety. Poll only after the worker
  reports done (the poll closes the task); gate on a scratch checkout
  via a temporary ref, never main; --after-runs takes the send's
  runs unchanged; no blanket pkill -f x2ssh.
• mu agent close no longer deletes cp -a workspaces silently. A
  none-backend workspace can't prove it is clean, so close refuses
  (WorkspacePreservedError) unless --discard-workspace. SKILL.md and
  waves.md now give the safe workspace free path and the refresh --from fix for unpushed merges.
• model examples with :<level> failed through a gateway
  provider. pi reads the first segment as the provider, so
  anthropic/…:high died with "No API key". The models recipe, delegate
  guide and mu_delegate help now say to copy <provider>/<model> from
  pi --list-models; a test pins that mu_delegate accepts
  provider/family/id:level.
• mu agent send to a dead pane no longer says "ctl: refused". When
  the control socket is silent and the pane is gone (the CLI exited,
  e.g. no API key), ctl verbs now raise AgentPaneDeadError with next
  steps to read the log, close the row and respawn, not link/kick/ssh
  advice. A live pane whose socket refuses still raises
  AgentCtlUnreachableError.

mu 3.10.2

The TUI exits on q again with sync on. Upgrade with
npm i -g @mu-crew/mu@3.10.2. No schema or pi extension change.

Fixed

• mu hung after q when MU_SYNC_DIR was set. The TUI stopped
  rendering but the process never exited: the 3.10.0 sync worker was
  unref'd before its message listener was attached, and attaching it
  re-refs the worker. It is now unref'd after its listeners.
• The TUI's exit flush no longer waits for the segment lock. After
  q, the flush could wait up to 2 s on a lock held by another mu
  process. The TUI (bare mu, mu state --tui) now tries the lock
  once; if it is busy, nothing is written and the ops stay in the DB
  for the next flush. The TUI's worker has already flushed every 10 s.
  Other verbs keep the 2 s wait. Exit is ~150-450 ms when the segment
  manifest is sealed. A mu older than 3.10.0 still running on this
  machine rewrites it unsealed, which makes every flush rescan the
  segment, so restart those processes.