git mv per the regrouping RFC: the five human-collaboration seams and
tui join packages/interaction/, app-boot becomes packages/boot/, and
jsonrpc joins the renamed scaffold/ (formerly sdk/) as its server half
beside client/protocol/create-sdk/helper/scripts/telemetry, whose
folders drop the legacy sdk- prefix. Three new group README triplets
replace the ui/ and sdk/ ones; tsconfig references/paths/globs,
knip keys, vitest globs, gate scripts, catalogs, docs, and the
lockfile follow. Adds the four settled FIXME rename markers
(dsh-sdk-server, dsh-sdk-telemetry, dsh-sdk-helper, dsh-sdk-scripts).
The scaffold folders diverge from their npm names until those renames
land, so tsconfig.base.json maps the three affected names explicitly
beside the group wildcard. Also repairs two pre-existing stale-path
classes the strengthened sweep surfaced: docs/web-styling.md's retired
web-ui host package and type-model spec fixture-literal joins.
app-boot's three Loader-composition specs time out at the default 5s
under full-suite parallel load on this filesystem (pre-existing;
pass isolated with --testTimeout=30000); interaction/scaffold/boot
suites otherwise green (687 passed).
ctx.subagents.interrupt() stops one live continuable child's current turn
via Agent.cancel(cause, { keepInbox: true }) under either a human durable
parent address or an exact live ancestor Agent. Fire-and-return: admission
is synchronous, quiescence is not awaited. Pending inbox work, the
Activation, and published descendants are preserved; only a later waking
send resumes the parked FIFO queue. Absent, one-shot, and disposing
targets are accepted no-ops.
The new Host RPC subagent.interrupt calls only that primitive with user
authority — no catalog, history, persistence, or parent-registry lookup —
so a live child stays stoppable while its parent Agent is offline.
Refs #1535
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.
The user-turn predicate (opensUserTurn) unifies the three parallel
consumers a new node kind silently missed — produced-files turn reset,
retry liveness, own-words force-scroll — so a skill invocation behaves as
the turn opener it is. The menu marker resolves through ctx.locale.bind
instead of a hand-rolled snapshot lookup; the dead legacy <skill> render
arm goes with the removal cut; command-over-skill name precedence is now
documented at the matchEnter seam; and the emptied replacement catalog
keeps the no-reload sentence, with the never-published residual recorded
in the Agent Note.
A menu pick or an entered /name line now claims the composer into an
args-tolerant skill.invoke transaction instead of shipping the literal
text and hoping the model loads the skill. This gives every user-invocable
skill a deterministic entry point — including disable-model-invocation
skills the catalog never shows the model (issue #1470). Candidates carry
a user-only hint, and the unreached legacy <skill> reference codec is
removed (decision 21 removal cut).
skill.list now serves every user-invocable skill and carries modelInvocable
so menus can mark user-only entries; the old model-and-user intersection
hid disable-model-invocation skills from their only legitimate entry point
(issue #1470). skill.invoke enforces user-invocation policy at the host
boundary, renders the canonical <skill_content> body, and injects it as a
user-role message carrying the skill-invocation source before starting a
turn. The connection fixture mirrors both faces for client tests.
A default naming a route the Models page has since removed left the
composer saying 选择模型 while the input still accepted a message, which
then failed inside the adapter mid-turn.
`session.prompt` now refuses with `model-unavailable` before opening a
turn. That is the enforcement boundary: the method stays callable no
matter what a client disables. `session.models` reports the same fact as
`routable`, and ui-model pushes a block through the new
`ctx.conversation.blocks` registry so the bar renders the disabled
textarea it already renders without a workspace, carrying the blocker's
own reason. The push direction is forced — ui-model already depends on
ui-conversation, so ui-conversation cannot read it back.
The gate is `routable`, not "matches no advertised group": catalog
membership is advisory, so a route serving a model it stopped advertising
is missing from the groups yet perfectly usable, and `null` before the
first load never blocks so a slow Host cannot lock a working composer.
The scaffold gains a route-only adapter for fixture-less keyless
scenarios. Registering zero providers is a test artifact — every product
composition mounts one — and the goldens that froze the seat's fallback
label now show the model those scenarios actually route to.
A row's stored profile could not tell a hand-declared gateway from a
shipped provider whose models someone narrowed — both look identical from
outside the adapter — so the Models page had no way to mark the routes a
deployment added itself.
The directory entry now carries `declared`, answered by the owning adapter
against its own installed catalog, and the page renders a Custom tag from
it. Absence stays "this adapter draws no such distinction" rather than
"shipped", so a route no adapter claims is labelled neither way.
Also records the default-route work's Agent Note and the e2e evidence for
all three changes: the composer switch writing the section, and the Models
page declaring a route with its own reasoning effort.
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.
The tool catalog boots every tool with its default config, but
allowParallelInProgress has no default, so the generator has to choose;
the page now says which branch it shows. The Agent Note's sidecar count
went stale twice inside this PR, so it records the refresh rule instead
of a point-in-time census. The SDK builtin entry pins its config literal
with satisfies like every sibling.
The client graph's ContentBlock narrows to the base four members, but
fixture turns carry merge-extended blocks at runtime, so removing the
JSON fallback crashed pricing on them. Keep the fallback and suppress
the narrowing-based lint finding with the reason inline.
Fixture-authored content is the closed base vocabulary, so after the
text/reasoning/tool-call arms only tool-result remains; the literal
comparison and its JSON fallback were dead and failed oxlint.
llm.discoverModels was reachable from any declared trusted host. The
method takes a caller-supplied baseURL and makes the host issue a GET to
it, then reports the status or the parsed body — so on a LAN deployment
an anonymous caller had a probe for whatever the host can reach and the
browser cannot, plus a path that carries a draft credential. The
PRIVILEGED_METHODS doc already states the rule this broke: trustedHosts
is a DNS-rebinding fence, not authentication, so the configuration plane
stays loopback-same-origin. It is in that set now, asserted both against
the hand-built fence and over real HTTP beside the catalog reads that
deliberately stay reachable.
supportsDiscovery and listModelDiscoveryNamespaces are gone. The field
was required on the wire and read by nobody: its own contract said a
surface should offer the action "instead of naming an adapter family it
would have to hardcode", while the surface hardcodes llm-pi-ai in two
places and gates the button on whether there is anything to probe. Its
shape did not fit the second caller either — the create card has no row
to read a per-row field from. Keeping a required field alive for a
consumer that may never arrive costs every producer and fixture a value
nobody consults, which is exactly how the fixtures drifted. The registry
that fed it had no other production consumer, so registration and
disposal are now observed through the offer itself.
The Agent Note claimed the key is never logged, which the wire schema
beside it already contradicts, and predated both the provider field and
the catalog-answer path. The two new public types pointed at core.md
without a type-equiv block or manifest entry, so the generated service
catalog named documentation that did not exist.
Clicking "fetch available models" on a built-in provider went to the
network. That is the wrong source: pi-ai's registry is the authoritative
list for its own providers, and it carries the context windows and output
caps a `GET /models` listing does not disclose. Asking api.deepseek.com
what DeepSeek serves is both slower and worse, and against an endpoint
that answers a different shape it failed outright.
Interrogation is still keyed by settings namespace — the provider being
added has no route — but the request may now name the route it is
editing. An adapter that already describes that route answers from what
it knows, needs no endpoint at all, and never touches the network; only a
route the catalog does not describe reaches the wire, and one naming no
endpoint is told to set one or enter its models by hand.
`ConfigurableProviderView` gained `supportsDiscovery` so a surface offers
the action where a namespace can answer instead of hardcoding an adapter
family.
Three narrower corrections ride along. Discovery no longer claims Azure
or Codex: Azure authenticates with an `api-key` header and an
`api-version` query despite its OpenAI lineage, and Codex uses OAuth, so
both reported an authentication failure as a provider with no models.
Cancellation during the body read escaped as the raw abort reason rather
than a coded ABORTED. And the schema comment claiming the probe key is
never logged overstated it: the host neither stores nor returns it, but
it rides the client's outgoing envelope like every other secret-bearing
payload, and redacting that tap is a configuration-plane-wide change.
Once a pi-ai route became a declaration rather than a catalog lookup,
adding an OpenAI-compatible gateway meant knowing its model ids up
front. Most such endpoints publish that list at `GET /models`, but no
seam operation could ask: every one is keyed by a registered provider
route, and the provider being added has no route, no stored profile,
and no stored credential — the endpoint and key are values in a form.
Interrogation is therefore keyed by settings namespace, which a
configuration surface already holds from the configurable-provider
directory. `registerModelDiscovery` offers it per namespace,
`discoverModels` asks, and the request carries the draft itself. The
reply is candidates, not a catalog: every field but the id is optional
because most listings disclose nothing else, and adopting one is a
settings write like any other. Nothing here reads or writes settings or
credentials, so `settings.yaml` still decides what a route serves.
`llm.discoverModels` carries the same draft over the wire. Its apiKey is
the third and last payload a secret may ride, and it is never stored,
logged, or echoed; every refusal folds into `model-discovery-failed`,
naming the endpoint asked but never the credential offered.
The pi-ai side is a plain GET for OpenAI-compatible protocols only —
their listing shape is the one gateways, self-hosted servers, and the
official endpoints agree on. Others say so, sending the user to
hand-entry rather than reporting a guessed shape as an empty provider.
The reply is read under a four-megabyte ceiling held on the bytes
actually received, because the endpoint is a URL the user typed.
The composer ring, percentage, and `~used / capacity` header read
`contextPressure.pressureTokens`, which moves only when a request reports
usage. Compaction reports none — compact-basic summarizes through a direct
`ctx.llm.stream()` call and appends only its own `compact/*` records plus the
replacement `user/message` — so the meter was frozen across the one action
taken to change it. Driving a real `compactNow` through the agent loop:
BEFORE compact: ring=4% header=~4227/100000 rows=[18, 0, 4365]
AFTER compact: ring=4% header=~4227/100000 rows=[18, 0, 286]
The composition rows fell 93%; the ring did not move, and would not until an
entire further turn completed. The panel then contradicted itself by more than
an order of magnitude at exactly the moment a reader opens it.
`contextPressure` now also publishes `projectedTokens`: the provider sample
plus the heuristic repricing of everything the surface gained or lost since
that sample, clamped at zero, folded through the shared `surface-fold.ts`. The
sample is stamped before the same event joins the surface, so an
`assistant/message` anchors against the surface its own request carried. Only
the delta is estimated, so the figure stays provider-anchored — the estimator's
CJK and JSON-schema underpricing stays out of the occupancy number — while
reacting the moment content lands or a span is shadowed. Same run after:
BEFORE compact: ring=4% header=~4323/100000 (pressure=4227, projected=4323)
AFTER compact: ring=0% header=~ 244/100000 (pressure=4227, projected= 244)
`contextOccupancy` prefers the projected figure and falls back to the bare
sample, so a projection restored from a pre-field checkpoint degrades to the
old behavior rather than disappearing. `stateVersion` moves to 3.
