host: the two 300-unit oneLine helpers cut on a code point boundary #1958

Implementation: FlyBot (ours) — PR follows within 24h. Do not pick up without a pickup-ok comment on this issue.

Written-by: FlyBot Lead
Claimed-by: FlyBot Lead
Claim-Id: 152521891c4c4c7cb0043e3094633caf
Branch: dev/1-unit1-oneline-code-point
Base: 37e8162
At: 2026-11-01T00:54:57Z

Implementation: FlyBot (ours) — PR follows within 24h. Do not pick up without a pickup-ok comment on this issue.

Written-by: FlyBot Lead

What happens today

Two independent 300-unit helpers still cut with String.prototype.slice, which counts UTF-16 code units. When the bound lands between the halves of a surrogate pair (an emoji or any astral character), the cut keeps the high surrogate alone; that lone surrogate is stored and every renderer shows it as U+FFFD.

• core/src/host/tasks.js:105 — oneLine(value, max); the 300-budget call sites are readNote (:124), :135, :161 and readTitle (:106). Base 37e81623 line numbers.
• core/src/host/work-dispatch.js:16 — a separate local helper of the same shape, .slice(0, 300), used at :33 to turn a group line into a dispatched card's title and note. Base 37e81623 line numbers.

They are two copies, not one shared function: fixing one does not fix the other, and a single test may not stand for both.

Reproduction (base 37e8162)

With only the two new test files added to a pristine 37e81623:

• core/test/tasks.test.js → a task note cut at the 300-unit bound keeps the pair whole fails with the stored note is not well formed: "a\ud83d".
• core/test/work-dispatch.test.js → a dispatch title cut at the 300-unit bound keeps the pair whole fails with the title is not well formed: "x\ud83d".

# pass 0 / # fail 2.

Expected

A bound that lands inside a pair drops the pair whole, so the stored string is always well formed, and the budget stays 300 units — it must not silently shrink to 299, and a pair that ends exactly on the bound must be kept.

core/src/shared/text-boundary.js already exports cut() for exactly this and core/src/host/tasks.js already imports it; no new helper is needed.

Scope

• core/src/host/tasks.js + core/test/tasks.test.js
• core/src/host/work-dispatch.js + core/test/work-dispatch.test.js
• the four patches/ds-bot/files/ mirrors, regenerated by generate on 37e81623 (never hand-edited), then cmp-checked against core/
• docs/subsystems/core-host.md: the sentence the arch-docs gate asks for, next to the existing 2,000-bound paragraph
• one changelog.d/ fragment
• one PR; each site has its own failing test first, and each of the two mutations (revert one site to slice) turns only its own test red

Out of scope

• Text already stored ill-formed by hand or by import: that needs a load-time toWellFormed in the store, a separate issue.
• The other .slice() sites in the tree (#1917, #1930, #1915 and friends covered their own sets; #1956/#1957 covered the 2,000-unit bounds inside tasks.js only).

Duplicate check (2026-11-01)

On live main 37e81623 both sites are still bare slice (read-only verification: 0 of the audited sites are already cut). Closed issues searched: code point, surrogate, oneLine, work-dispatch 300. The closest are #1917 (groups/browser/diagnostics/remote-label), #1930 (busy-ack/delivery/tools), #1956/#1957 (the 2,000-unit bounds in tasks.js); none touches tasks.js:105's 300 budget or work-dispatch.js:16. No duplicate, no follow-up issue pending.