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.
pi-ai resolves a request's apiKey override only through a provider that
declares an api-key method: resolveProviderAuth short-circuits to that
method when the override is present, and otherwise falls through to the
credential store and then to ambient discovery. A provider with no
api-key method at all therefore resolves to nothing, and the request
fails with "Provider is not configured" before any network I/O.
Two routes hit that. openai-codex ships OAuth alone, so moving off the
/compat dispatch broke a profile that names a key for it — the old path
handed the token straight to the provider. And a catalog route naming an
api was being rebuilt with the harness's own auth, so `openai: {api:
openai-completions}` stopped reading OPENAI_API_KEY, contradicting the
documented promise that omitting a credential keeps provider-native
discovery.
Auth is now one decision for both constructions. A catalog route keeps
its installed provider's auth, through an api override too: which
environment a provider reads belongs to the provider, not to the wire
format its models speak. A catalog provider with no api-key method gets
the harness method beside its own, but only when the profile names a
credential — a keyless codex profile keeps the honest refusal, since
this adapter holds no OAuth store to resolve through.
Materialization now spreads the installed entry instead of enumerating
the result, so a Model field this package does not model survives a
pi-ai upgrade; headers went missing from an nvidia route exactly that
way once already. providerInfo reports the configured displayName, which
also joins the registration facts so a rename re-registers rather than
leaving the old label in every selector. A refused registration swap
gets its own diagnostic naming the route, matching the directory swap
beside it.
The README documented endpoint interrogation this layer does not
implement, and still described unknown providers as kept-last-good after
they became legal declarations refused at the write point. The Agent
Note claimed per-model reasoning configurability the schema never had,
required capacities the route now defaults, and stated an apiKey
override that short-circuits unconditionally.
The adapter omits the seam's reasoning field whenever a model carries no
reasoning metadata, which is the model's own property and says nothing
about where the model came from. Both the JSDoc and the Agent Note read
as though only hand-declared models were meant, so a reader would infer
that the 251 installed-catalog models pi-ai marks as non-reasoning still
offer their single off level. They do not, and that is the point: a
picker holding only off misrepresents a provider that thinks by default,
because off dispatches the same bytes as naming no effort at all.
Behavior is unchanged; only the prose that describes it was narrower
than the contract. adapter.spec.ts already pins the catalog case through
openai/gpt-4.1 and catalog.spec.ts pins the hand-declared one.
Review found four defects in the declared-provider work.
`PiAiAdapter` reused one `Models` collection and mutated it whenever the
configuration changed. `Models.streamSimple()` resolves its provider
lazily — when the stream is first consumed, which is after the adapter
awaits the route's credential — so a configuration change landing in
that window let an in-flight request finish under a configuration it
never resolved against, or fail on a provider that no longer existed.
Each resolution now produces an immutable snapshot and every operation
captures one before its first await, which is what makes the seam's
per-step freeze (`llm.prepareCall()`) hold end to end: switching models
mid-reply takes effect on the next step, never inside the one in flight.
`defaultMaxTokens` was materialized from the catalog's `Model.maxTokens`.
The two answer different questions: pi-ai requires that field as the
model's output capability, while the seam's is a cap the deployment
chose to send on requests naming none, so every request had started
carrying a number nobody picked. Only an explicitly configured cap
reaches the seam now.
The configurable-provider directory was refreshed by disposing its
registration and making a new one. A candidate set the registry refuses
— a profile keyed `deepseek-official`, which llm-deepseek declares —
left the whole directory withdrawn and the Models page empty, silently,
because the settings callback contains the failure. The seam's
registration handle now carries `replace()` with the same
validate-first atomicity `registerAdapter` has.
The protocol table offered every pi-ai streaming API, including four
whose authentication a profile cannot express: Bedrock signs with SigV4
over AWS credentials and a region, Vertex needs a project, a location,
and ADC, Azure needs provider environment plus an api-version, and Codex
uses OAuth. Offering them handed back routes that cannot authenticate.
Catalog routes still reach them through their own provider.
A pi-ai route had to name an installed catalog provider, served that
catalog's models verbatim, and could override only the endpoint. An
OpenAI-compatible gateway, a self-hosted server, or a model newer than
the pinned pi-ai release was therefore unreachable, and a stale context
window could not be corrected without upgrading the package.
A route is now a declaration whose defaults come from the installed
catalog. `catalog.ts` merges that catalog under the profile's own model
entries, `provider.ts` builds the pi-ai Provider (reusing the catalog
provider when the route keeps its protocol, so implementations this
package cannot reconstruct keep working), and the adapter serves every
operation from one `createModels()` collection. That also retires the
`@earendil-works/pi-ai/compat` import, which pi-ai documents as a
temporary entry point it deletes with its ModelManager migration.
Credentials stay on the harness seam: the resolved key rides the request
as pi-ai's highest-priority auth override, so `Models` holds no
credential store and a named-but-missing reference still fails loud
instead of falling back to an unrelated ambient key.
A model's configured maxTokens now reaches the seam as defaultMaxTokens.
- docs: fix cross-class note links (../feature/...), mirror the membership
reuse criterion in README.zh.md and the connectWorkspace architecture
note (EN/ZH/sidecars re-recorded)
- test(client): stray + member blank at the same path — the scan skips the
stray and reuses the member
- test(web): re-registration after delete now mints a fresh blank session
(the old cwd-only reuse is gone), so the account is no longer empty;
assert the retained seeded session is still not re-adopted
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 hover affordance framed only the strip right of the crumbs. The bar itself
now carries the outline and the padding in both modes, so hovering previews
exactly the field the click produces and nothing resizes when the two swap.
One keystroke moved the view twice: deleting a separator first narrowed the
pane the draft had just walked away from, then replaced it with its landing.
The tail now filters only the LAST pane — the one whose level the path names —
so a pane on its way out holds still until its landing arrives.
Also from the review round: the walk waits both legs out instead of taking the
submitted-navigation bound (a speculative scan has nothing waiting on it, and
a tail keystroke aborting a slow parent leg would otherwise strand the
two-pane view); a level keeps answering the directory text that produced it,
so `..` segments and Windows forward slashes filter and stop rescanning; the
release-on-miss rule counts displayable rows, so it survives `hidden` ever
meaning more than dot-prefixed; and the editor's 250ms rest joins the other
two constants on the remote-recalibration list.
Skipping the draft-following scan whenever ANY pane happened to list the
directory was the cheaper rule and the wrong one: erasing a segment left the
level being typed on the LEFT, with its own child pane still standing to its
right, so the two panes stopped reading as "where I am, and where I came
from".
The pane arity is now the invariant the editor maintains — the last pane lists
the level the path names, its parent sits beside it, and only a display root
lists alone. Only that last pane's own tail costs no scan; every other
directory part re-lands.
The draft-following scan replaced the panes with one wide level, so typing a
path collapsed the dialog's Miller view — the thing the dialog is. It now
lands through the same selection-anchored landing every navigation uses:
target and parent legs as one frame, the target re-selected in its parent
level, its children on the right. Typing a path moves the Miller view exactly
as a crumb jump does.
One landing shape, two callers: `land(path, {closeEditor, announce})` is what
`navigate` and the draft-following scan share. A submitted path closes the
editor and announces failures; the speculative scan keeps both to itself and
re-parks the focus its swap dropped.
A level a pane already lists still needs no scan at all — the filter alone
answers the draft — so erasing back into the parent's own path keeps both
panes and only moves the filter.
ds-review-bot round one. Keying the debounce on the directory part the draft
named left two states with no recovery until the operator crossed a separator:
a keystroke that superseded an in-flight scan never re-armed one, and an edit
after a rejected submission released the hold with no timer left to release.
The wait is now keyed on the draft itself and decides its target when it
fires, reading the panes through a ref so a landing cannot re-arm it (a host
answering with a differently spelled path would otherwise scan forever).
A landed scan that unmounts the row a keyboard operator Tabbed onto re-parks
focus on the still-open editor; the Modal has no focus trap. That a walked-to
level survives closing the editor is now stated in the README, the Agent Note,
and the module contract. The new e2e stages its own beta directory so running
it alone sees the tree its assertions describe.
The Select Workspace Directory dialog hid its one route into typing a path
behind an invisible click target, and once the editor opened the panes stayed
on whatever level was listed when it opened — so the typed text and the list
under it disagreed for the whole edit.
The edit zone now carries a pencil glyph at the bar's right edge and lights in
the editor's own footprint on hover/focus (the bar keeps one height across the
swap). While editing, the panes follow the draft: a directory part no pane
lists is scanned after a 250ms rest and lands in place, so typing deeper
descends and erasing segments steps back up without leaving the editor, and a
final segment nobody matches releases the prefix filter instead of emptying
the pane it is being spelled into. The draft-following scan is speculative and
silent on failure; Enter still owns the view from submission until landing and
remains the only path that surfaces an error.
Agent setup may await while a mutable contribution registry changes. The previous subagent path validated and committed its provisioning batch inside the setup callback. A revocation queued after that callback returned therefore treated the installation as resident and released it, even though AgentLoop had not published the child yet. AgentLoop could then admit and announce a child whose required capability had already disappeared.
Introduce AgentSetupCommit as the optional synchronous result of create and resume setup. AgentLoop now awaits setup, invokes that commit with no intervening asynchronous boundary, and only then enters the Session and Agent registries. A commit failure follows the existing private-transaction rollback, so neither identity is published and the caller can reuse the id.
Keep continuable-subagent installations provisional until this publication commit. Contribution removal still releases every installation immediately, but now marks an unpublished batch invalid so its commit rejects with ACTIVATION_SETUP_REVOKED. Once the commit succeeds, later removal remains ordinary live revocation.
Cover create and resume ordering, resume commit rejection and identity reuse, and an assembled microtask revocation that leaves only the parent Agent and Session. Update the public JSDoc, architecture flow, package contracts, current Agent Notes, Chinese counterparts, pairing records, and generated Cordis API to describe the new boundary.
Validated with the four focused Agent/subagent test files (91 tests), the isolated assembled regression, targeted TypeScript project builds, generated Cordis API freshness, export JSDoc verification, scoped translation pairing, Markdown wrapping, and Mermaid parsing.
ds-review-bot round 1 on the repository-plugin runtime:
- a manifest-declared skill root absent or non-directory in the installed
package now fails the plugin load (skill-local treats a missing root as
legitimately empty, which silently mounted a skill-less plugin)
- includeDefaultRoots: false no longer inherits $DSH_BUNDLED_SKILL_DIR, so
isolated repository providers see only their explicit roots
- prepared wrapper baseUrl schema requires the file: scheme, failing hostile
URLs at the declared validation boundary
- preparedPath reuses format.ts's isOutside; SERVER_NAME_PATTERN is exported
and pinned equal to dsh-mcp-client's, with the restatement justified (the
prepare bin keeps a zod-only module graph); the unexplained `as never`
cast now carries its schemastery rationale
- the import-free wrapper assertion also rejects dynamic import(
- the headless fixture wrapper is regenerated by the real prepareDshPlugin
and a drift test pins fixture == generator output
- prepareDshPlugin JSDoc states the non-atomic publish repair contract
- inline the collect() identity wrapper now that both streams use the
seam's diagnostic-tail shape
- resolve the packaged rg path lazily at the first call (memoized):
@vscode/ripgrep resolves its platform package at module evaluation, so a
static import turned a missing/corrupt platform package into a Loader
composition failure instead of the documented per-call SEARCH_FAILED
- classify synchronous spawn-creation throws (a NUL in argv, an abort
racing the pre-check, a rejected resolution) into SEARCH_FAILED /
SEARCH_ABORTED instead of leaking raw errors
- correct the stderrMaxBytes contract: the stderr excerpt is embedded in
SEARCH_* error messages, not hidden from the model
- export virtualManifest and pin its three acceptance paths (prefix hit,
pnpm-11 truncated-name content-scan fallback, both miss) with fixture
unit tests
Tests: rg-path.spec.ts (resolution failure + memoized rejection),
tools.spec.ts spawn-creation classification, notices spec virtualManifest.
Merges master (default-workspace-write-ui, installer-adopt-checkout,
remove-scoped-bash, goal-clear-single-flight, frontend-plugin-loader,
install-interface-choice, …).
Conflict resolutions:
- apps/cli/tests/shipped-composition.e2e.ts: keep the fixed glob/grep
roster assertion; master's workspace-write composition now confines
tool-bash, so the escalation pair is pinned present (my earlier
absence pin is superseded) together with master's permission facts.
- even-out-shipped-tool-rosters note: both sides edited it; the merged
text keeps both sets of changes and the pair is re-recorded.
