The task registry has run every background bash, pwsh, pty-send, and
one-shot subagent since it landed, but only the model could read it: a
human at the Web client could not see that a build was running, tell a
finished task from a stuck one, or find its outcome anywhere but the
`run_in_background` tool card that printed an id and never updated.
Task state now reaches the browser as one whole-snapshot `session/tasks`
mux frame per session, pushed at every registry commit that changes what
that session can see. `TaskService` gains `onTasksChanged`, which is
owner-granular because owner-disposal removal is a change no per-task
record can express. The carrier reads the exact owner the listener hands
it, so a push stays correct while that scope tears down, and reads the
baseline through the non-resuming `ctx.agents.get` so listing never
revives a cold session. The client keeps a last-wins mirror on
`SessionListState`, and a new `dsh-client-ui-task` package renders it
beside the subagent catalog — rendering nothing at all until the session
has a task, so an ordinary conversation grows no new chrome.
Streamed per-task output and human-initiated cancellation are separate
phases; the note records why neither has to undo this channel, and why
no Web path may call the consuming `ctx.tasks.read()`.
Invocation is an ordinary session.prompt again: the pre-step gesture
boundary makes it deterministic host-side for every front end, so the
dedicated RPC (handler, wire schema, error codes, client face, fixtures)
and ui-skill's claim machinery are net deletions. The menu keeps decision
21 exactly — a pick lands literal /name text — plus the user-only marker
from skill.list's modelInvocable flag.
Review findings on #1977, each verified before acting: the 2026-08-03
declared-provider-catalog note is updated in place and cross-linked both
ways now that reasoningEfforts/compat reopened half of its rejected
alternative; resolveModelCompat inherits the catalog entry's compat only
while the resolved api still is the entry's own, so a route-level api
repoint no longer merges another protocol's shape as a completions base;
the off-with-value promise gains a request-boundary test proving pi-ai
reads thinkingLevelMap.off when the reasoning option is absent (and the
catalog-level test name stops overclaiming); the cannot-stop-thinking
wording narrows to what is actually enforced (no Off offered, explicit
Off refused — an effortless request goes out bare); the z.const(null)
comment attributes null passthrough to schemastery's nullable
short-circuit; the baseten drift-gate claim names its verification
source; and the layered-merge delete gap for dict keys is documented
under Known Limitations with the atomic-leaf follow-up in #2003.
Bilingual README updates for the four touched packages (ui-skill's claim
flow and deterministic-injection model experience, the apiproxy skills
domain, the shared renderSkillContent seam export, the catalog stitch
sentence), the implemented Agent Note triplet recording the decision and
its peer-product evidence, and the regenerated catalogs/graphs.
The user guide's model-catalog section teaches the three new knobs at
task altitude — declare levels per model, pick the reasoning dialect,
reshape catalog models with modelOverrides — with the settings.yaml
example exercising all of them, plus an UNSUPPORTED_REASONING_EFFORT
troubleshooting row. The generated plugin config catalog picks up the
new Config fields, and the bilingual Agent Note records the decision,
the alternatives considered, and the schemastery materialization
constraint that chose false over {} as the disable spelling.
Yichen Jiang
imccyu
Tianyi Cui
Turtle