[lawrencecchen]

What

Per-token copies removed on the claude-stdio path (what Lawrence's chats use):

• Agent host: each entry is encoded once, from a borrowed view, and kept as its frame bytes. The live send and a resume send the same bytes. Before, each entry was serialized once to weigh it, deep-copied to send it, and serialized again by the writer.
• Message::from_value moves the fields out instead of copying params. New Message::into_value; the host builds records with it. The raw Claude line moves into its record. It is now translated first, and the raw record still comes first.
• Message::to_line serializes from the borrowed fields. Before, to_value copied every param for each client line.
• The preview grows in place instead of being copied on every chunk.

Numbers (rbx, perf100 bench: 10 sessions x 4,000 deltas, one attached client; base 26e79e1; two runs each)

Metric	Before	After	Ratio
Agent host CPU per token (release)	27–28 µs	13–14 µs	2.0x
Daemon CPU per token (release)	89.5–93.5 µs	75.5–85.7 µs	1.1–1.2x
Agent host CPU per token (debug cargo test)	33–35 µs	23 µs	1.5x
Agent host RSS after a 20,000-delta turn (debug)	64–89 MB	36–40 MB	~2x
Log bytes / write syscalls per token	674 B / 2.35	same	1x

Most of the daemon's per-token cost is still in the hub: two records per token, the log line, the event ring's record copy, and the fan-out. That is the next slice; it needs a profiler on rbx (perf is blocked for the rbx user).

Test

tests/claude_stream_budget.rs: a recorded-shape turn of 20,000 deltas from a stand-in Claude Code (tests/fake_claude_stream.py), run through the CLI. It checks that the reply arrives whole. It also checks that the agent host uses at most 70 µs and the daemon at most 330 µs of CPU per delta, about 3x the measured debug values. It is a budget guard: on base the debug numbers (host 33–35 µs, daemon 129–136 µs) also pass.

Review

A correctness review subagent checked this change:

• The encoded bytes are identical to write_frame(&HostFrame::Entry), and the controller decodes them.
• from_value gives the same result for every input.
• The raw record still comes before the translated records and the stdin taps.
• The writer still does write_all and flush per frame.

Nits that remain open:

• An entry over the frame limit now becomes a stderr note. If it was a response with an id, that request is not answered. Before, the controller was dropped on every resume. Input lines are capped below the limit, so this is hard to reach.
• The buffer weight now counts the frame overhead, so the stderr drop threshold triggers slightly earlier.

Changelog

none

Observed / Not run

• Observed on rbx: the bench, the budget test on head and on base, and cargo clippy -p acpmux --all-targets -D warnings, where the only errors are the base reds cmp_owned (cmux-tui: two base clippy reds (acpmux steer_now_tests, worktree-care store) #22906) and unnecessary_sort_by in cmux-worktree-care. The full suite runs in the landing gate.
• Not run: macOS. The acp-fork lane's Codex bench was not rerun; the Codex bridge shares from_value and into_value.

---

^{Need help on this PR? Tag @codesmith-bot with what you need.}

---

Summary by cubic

Removes per-token copies on the claude-stdio path. The agent host encodes each streamed entry once and keeps its frame bytes, sending the same bytes live and on resume; Message::from_value and the new Message::into_value move fields instead of copying params; Message::to_line serializes borrowed fields; and the streaming preview grows in place. Agent host CPU per token drops from 27–28 to 13–14 µs (release, perf100 bench) and RSS after a 20,000-delta turn from 64–89 to 36–40 MB; daemon CPU per token drops from ~90 to ~78 µs.

Budget test

• tests/claude_stream_budget.rs runs a recorded-shape 20,000-delta turn through a stand-in Claude Code (tests/fake_claude_stream.py) and asserts the reply arrives whole and that host/daemon CPU per delta stay under 3x the measured debug values.
• An entry over the frame limit now becomes a stderr note instead of dropping the controller on resume, and the stderr drop threshold triggers slightly earlier because the buffer weight counts frame overhead.

^{Written for commit 7c1e67d. Summary will update on new commits.}

[github-actions]

cmux status _{(one comment, edited in place; each line is updated by the job that owns it)}

Review 7c1e67d: HQ feed card · gallery before/after runs · checks

App of this head (app.zip, kept 1 day)

gh workflow run reload-build.yml -R manaflow-ai/cmux --ref feat-cmux-next -f tag=pr23051 -f ref=perf100/acpmux-token-copies -f platform=macos -f backend_mode=local

[lawrencecchen]

Correctness review (independent subagent, correctness first) and resolution

The review covered #23052 (purge guard and cleanup) and #23051 (per-token copies).

#	Finding	Severity	Resolution
1	A chat that moves A → B → A is purged on A before its import, so the purge guard refused its save and records, and a restart of A lost it	blocker	Fixed in fda6f6e: transfer_in calls Store::release_purge(id) after it purges the moved-away copy. Red-then-green test a_chat_moved_away_and_back_survives_a_restart_of_its_first_daemon
2	LocalStore::save held the guard lock through its synced write, so every append in the daemon waited	should-fix	Fixed: the save checks before and after the write; a purge during the write deletes what it wrote
3	MemoryStore was not the same as LocalStore (append had no check; save checked without the lock)	nit	Fixed: both check under the store's own lock
4	One unreadable events dir stopped the whole startup cleanup	nit	Fixed: an unreadable log counts as a log, so the session is kept
5	An oversized In response now becomes a stderr note, and its request gets no answer (#23051)	nit	Open (hard to reach: input lines are capped below the frame limit). Follow-up
6	The buffer weight now counts the full frame (#23051)	nit	Accepted: the threshold triggers slightly earlier and stays harmless
7	Small copies remain (write_claude_line, error parse)	nit	Accepted

Checked and fine:

• Deadlock: writers → purged in both append and purge; delete holds no lock during its retry sleep.
• Startup order: daemon.lock is taken before load_from_store.
• The cleanup deletes only directories with status Closed and no log bytes.
• Fork and adopt use new ids; peers do not write to the store.
• The guard set stays small: about 50 bytes per purge per daemon run.
• acpmux: one serialization per streamed entry; agent host 27 -> 13 us/token (perf100) #23051: the frame bytes are identical to write_frame(&HostFrame::Entry); from_value gives the same result for every input; the order of records is kept.

Risk tier: #23052 adds a delete of user data, which is risk:high in REVIEW policy; this repo has no risk:high label. Lawrence chose the deletion (via the perf100 coordinator, 2026-10-21): "delete the 7,885 leftover records".

[teamleaderleo]

Bead cx-5quep created for this PR by the cmux coordinator.

[teamleaderleo]

cmux-next guards / defaults-snapshot is red on 24 of 24 fresh feat-cmux-next PR heads in the last 2:00:00, so it is most likely a base red, not this PR's. The base-reds lane has it; a base merge picks up the fix once it lands.

[cursor]

Bugbot needs on-demand usage enabled

Bugbot uses usage-based billing for this team and requires on-demand usage to be enabled.

A team admin can enable on-demand usage in the Cursor dashboard.

[lawrencecchen]

Rebased onto feat-cmux-next 390c286 (head 7c1e67d) with no conflicts. Note for the captain: since #23203 the hub no longer logs raw claude.stream_event lines, so the daemon's per-token numbers in the description (measured before it) are higher than on the current tip; the agent-host change (single encode per entry) is unaffected. Not re-gated after the rebase.

[blacksmith-sh]

Found 2 test failures on Blacksmith runners:

Failures

Test	View Logs
nightly cmux-tui resolve/nightly cmux-tui resolve	View Logs
test_the_repository_highlights_validate (main.DigestTests.test_the_repository_highl ights_validate)/ test_the_repository_highlights_validate (main.DigestTests.test_the_repository_highl ights_validate)	View Logs

^{Need help on this PR? Tag @codesmith-bot with what you need.}

[lawrencecchen]

Closed by the perf lead (agent, cmux-ad order 2026-10-31): stale, with no owner. This is an agent PR (lawrencecchen account). It was opened on 2026-10-21, its last commit is from 2026-10-22, no person pushed or commented in the last 3 days (automated review and sweep notes do not count), no live session names it in the lead status of the last 24 h, and it conflicts with feat-cmux-next (DIRTY).

The branch perf100/acpmux-token-copies stays. To reopen: click Reopen (or gh pr reopen 23051 -R manaflow-ai/cmux), merge origin/feat-cmux-next into the branch, get a review receipt on the new head, then add ready-to-land.