Three things the page got wrong.
The web-search provider's defaults lived only at their use site, so the
served section carried no value for them and the card fell back to a zero it
invented — a number the schema itself rejects. Declaring them on the schema
makes the settings service the one authority: `maxUses` now reads 5 because
that is what the Host resolves, not because the page guessed. `baseURL` keeps
its code-side default, which exists so `$DEEPSEEK_SEARCH_BASE_URL` can win.
A field the Host serves no value for now renders empty rather than as zero.
The cards were rows in a settings page of cards: name and description ran
together on one line because the shared disclosure row lays them side by side.
Each card now draws its own header, stacking the two, and the section follows
the idiom the Agent Preset page established.
Three scenarios over the real wire: the cards this deployment's exposed
namespaces produce, one field edited down to the settings document with the
override badge that layering produces, and the reset that clears it back to
the composed default.
Every other settings golden gains the same nav entry and nothing else.
The composer's preset seat stays removed here — this layer moved it to the
hero chip and the session-header action — so only the model seat takes the
`modelSeatLocked` narrowing master introduced.
Conflicts:
apps/web/tests/snapshots/*/*.expected.md
packages/client/connection/README.md
packages/client/connection/README.zh.md
packages/client/ui-conversation/src/client/skeleton/InputBar.tsx
packages/host/apiproxy/src/fetch/client.ts
**Reasoning effort leaves the provider cards entirely.** It is a per-MODEL
capability and the models under one provider disagree about which levels
they accept: setting `anthropic` to `max` made six of its eight models
throw UNSUPPORTED_REASONING_EFFORT, and because the catalog build catches
per provider, the whole provider vanished from the picker behind one error
row. A provider-scoped control can only ever be set to a value some of its
models reject.
The composer's model picker already offers each model its own levels, and
a switch there now records provider, model, and effort together as the
next session's default — so the setting has a better home at the right
granularity. The profile field stays in `settings.yaml` for a deployment
that knows its route; only the control is gone, from both cards and both
adapter families. Two `components.spec` cases used the control as the
vehicle for their op assertions and now use `baseURL`, which is what they
were actually testing.
**A rejected Provider ID now reads as a fault.** It shared the neutral
hint paragraph with the field's guidance, so the copy telling the user
what they got wrong looked like advice. Reuses the existing `.error`
style, matching the split the key field already makes.
The browser e2e lane had been failing wholesale since this stack moved the
agent plane into presets, and nothing caught it: 34 of 48 files. Two of the
causes are product defects, not test breakage.
`bashEnv` goes back to the host plane. `apps/cli/src/web.ts` injects it to
publish `DSH_WEB_URL`/`DSH_WEB_MODE`, so the earlier note that "nothing outside
the agent plane injects bashEnv" was simply wrong — behind a preset's `shell`
realm those variables reached no shell at all, and a `dsh web` agent could not
find the address of its own interface. This is the same criterion that returned
`subagents`: a host row that injects a service resolves before any session
exists and has no agent to key by, so the service is host-plane. `tool-bash`
consumes the host registry from inside the preset, which works because an
agent context chains to the host; only the reverse is invisible.
`tool-subagent-report` goes back with it. It is not a tool this agent calls: it
registers a continuable SETUP on the host `subagents` singleton, and that list
is not scope-aware. One copy per mounted preset meant every child was handed
`report` once per live session, so the second registration threw and a cold
subagent resume failed with `subagent-not-resumable` — a diagnostic three
layers removed from the cause.
The lane's own composition facts follow. Skill roots resolve inside a preset
now, a subtree include patches cannot reach, so the scaffold pins the roots'
documented environment fallback for its whole lifetime rather than for the boot
— presets mount per session. Without it the developer's real `~/.dsh/skills`
enters replay requests and goldens while CI sees none. The `apps/cli`
composition test pins `storage-json` for the same reason: unpinned it wrote,
and then read back, the developer's own `~/.dsh/storages/`.
Three tests now address through an agent what they used to read off the root
context, because that is where the thing lives: the tool catalog, the skill
registry, and the token meter. The seeded-history projection baseline asserts
the opposite of what it did — a detached session yields a preset-plane
projection only from a durable checkpoint written while it was live, and this
seed was written straight to persistence and never ran.
Goldens re-recorded for the hero's preset chip and the settings nav entry.
The Models page could name a provider's key and little else. Adding an
OpenAI-compatible gateway meant opening $DSH_HOME/settings.yaml and
knowing the profile shape; correcting a stale context window meant the
same. This layer puts both on the page: a card that declares a route
pi-ai does not ship — id, endpoint, protocol, key, models — and a model
list on the pi-ai editor that can ask the provider what it serves and
adopt the answer.
It follows the DeepSeek catalog editor that landed in #1050 rather than
inventing a second look for the same job. Both editors now share the
section shell and heading, the danger-tinted delete, the add-model
button, the empty state, the per-row validator that names a bad row by
its position, and one K/M capacity vocabulary — 256K and 1M are read and
spelled back, while settings.yaml still stores plain token counts. The
row type is structurally open like that editor's, so a profile field
this card does not edit survives an edit here.
Three of that editor's decisions replaced weaker ones this branch had
made. Inheritance now reads the composition base rather than the
effective value, which would echo an override back the moment a reset
dropped it. Validation names the offending row instead of stating a
blanket problem. And emptying the list is no longer conflated with
handing the catalog back to the adapter — those are separate acts, with
separate affordances.
The create write carries the revision the card opened at, so a route
another tab declared meanwhile is a conflict rather than a silent
overwrite of its profile.
Both editors live in `ui-models` and list the same thing, so they now share
one row shape rather than diverging when #1368 lands: a bordered entry per
model, id and display name on the row, and the capacities behind the row's
own disclosure. The context window is joined there by the per-model output
cap the adapter just gained; both read a decimal K/M suffix.
The shared class names carry this file's token spellings, not that branch's.
`--dsw-alias-border-subtle`, `--dsw-alias-text-tertiary`, and
`--dsw-alias-text-primary` are undeclared, so they resolve to the light-mode
literals in their fallback slots — the defect this section was moved off. A
styles test now rejects any `--dsw-*` name the token sheet does not declare,
so the next editor to name one fails instead of shipping a light-only
surface.
The keystroke buffer is now per capacity field rather than per row, since a
row holds two of them.
The catalog's context window is now a text field that reads a decimal K or M
suffix — 1M is 1000K, matching how model capacities are quoted — and stores
the plain token count, so settings.yaml and the adapter are unchanged.
A stored count reads back in the shortest form that round-trips: 1000000 as
1M, 256000 as 256K, and 131072 written out, because it is not a whole number
of thousands. The field holds the typed text while its row has focus, since
re-deriving it from the parsed count on every keystroke would rewrite 1000
to 1K mid-word; text that does not parse stays on screen so the save-time
rejection names a row the user can still see and correct.
