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left the whole directory withdrawn and the Models page empty, silently,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.
A catalog entry may now carry its own `maxTokens`, matching the shape the
pi-ai adapter already exposes. Exact-model resolution prefers it over the
profile value, so capping one model no longer means capping the route.
Defaults are unchanged: an entry without a cap, and any unlisted
pass-through id, still resolve to the profile `maxTokens` (256,000), and the
shipped catalog keeps its context windows.
Resuming a long session (196k events, 2.2k steps, 1.8k tool cards) took
~12s to render and ~800ms to echo one keystroke:
- Every step's timing footer called stepTimingAt, which replayed the whole
event log per footer - O(steps x events) on the initial render.
- pi-tui re-renders every component each frame and relies on per-component
line caches, but ToolCardComponent/ContextCardComponent built throwaway
Text/Markdown instances inside render(width), re-wrapping every settled
card's output on every keystroke.
Replace the per-footer replay with one shared StepTimingTracker per chat
mount (single O(events) cursor over the append-only log), and cache card
rows by width via CardLineCache, dropped by every state mutator and
invalidate().
Measured (tmux 200x50, 196k-event session): resume prompt-ready ~12s -> ~7.6s;
per-keystroke echo ~800ms median -> ~11ms.
The ShellDialect marker on BashExecutor and the load-time rejection in
tool-bash/tool-pwsh force every test and example fake executor to declare
a dialect for a guard with no in-repo or plausible deployment to catch
(shipped compositions always pair the tool with its matching executor),
and they break the example composition suites that stub ctx.bash without
a dialect (agent-spine-demo/cli-demo/acp-demo lost the bash tool).
Keep the non-dialect doc fixes that landed with the attempt: the tool-bash
README bashEnv contract trim with the stale example import removed, the
acp-snapshot suite.ts 'usable pwsh' JSDoc, and the pwsh.cordis.yml comment
indent. The parity note records the attempt and revert under Alternatives.
The seam gains ShellDialect ('bash' | 'powershell' - concrete shells, not
families: zsh or fish would be their own values, never 'bash'); bash-local
declares bash (bash-sandbox inherits), pwsh-local declares powershell, and
both tools throw at load when the mounted executor speaks another dialect -
previously tool-pwsh over bash-local handed PowerShell text to bash -c and
the deployment error surfaced as ordinary nonzero exits. Pinned by mismatch
tests on both tools; the parity note records the contract (both languages).
Also from the review round: the tool-bash README's managed-environment
section becomes a summary linking the owning dsh-bash-env contract (the
duplicated prose carried a stale owner in its example import), the
pwshOnly JSDoc drops the stale 'on PATH' phrasing, and the task-tools
contract comment in the two pwsh compositions is indented into its block.
