@sayknow-cli/coding-agent 0.5.26 → 0.6.1

package/src/config/model-profiles.ts

@@ -141,6 +141,21 @@ export const BUILTIN_MODEL_PROFILES: readonly ModelProfileDefinition[] = [
 		critic: "anthropic/claude-opus-5:high",
 		architect: "anthropic/claude-opus-5:xhigh",
 	}),
+	/**
+	 * Opus 5.5 is a separate preset instead of a bump of `claude-opus`, because its
+	 * envelope differs: it costs less than Opus 5 ($4/$20 vs $5/$25) at the same
+	 * 1M context / 128K output, and it accepts the `max` adaptive level. So the deep
+	 * lanes are pushed one notch above the Opus 5 preset (`architect` at `max`,
+	 * `planner` at `medium` instead of `low`) while the mechanical executor lane
+	 * stays on Sonnet 5. Anyone who wants the old cost/effort shape keeps `claude-opus`.
+	 */
+	profile("claude-opus-5-5", ["anthropic"], {
+		default: "anthropic/claude-opus-5-5:xhigh",
+		executor: "anthropic/claude-sonnet-5",
+		planner: "anthropic/claude-opus-5-5:medium",
+		critic: "anthropic/claude-opus-5-5:high",
+		architect: "anthropic/claude-opus-5-5:max",
+	}),
 	profile("claude-fable", ["anthropic"], {
 		default: "anthropic/claude-fable-5:xhigh",
 		executor: "anthropic/claude-sonnet-5",
@@ -402,6 +417,7 @@ const PROFILE_PRESENTATION: Record<string, ModelProfilePresentation> = {
 	},
 	opencodego: { displayName: "OpenCodeGo", providerGroup: "OPENCODEGO" },
 	"claude-opus": { displayName: "Claude Opus", providerGroup: "CLAUDE" },
+	"claude-opus-5-5": { displayName: "Claude Opus 5.5", providerGroup: "CLAUDE" },
 	"claude-fable": { displayName: "Claude Fable", providerGroup: "CLAUDE" },
 	"glm-eco": { displayName: "GLM Eco", providerGroup: "GLM" },
 	"glm-medium": { displayName: "GLM Medium", providerGroup: "GLM" },
@@ -460,7 +476,9 @@ const MACOS_OMLX_PROFILE_RANK = new Map<string, number>(MACOS_OMLX_PROFILE_ORDER
 
 const PROFILE_RECOMMENDATIONS: Record<string, string> = {
 	"openai-codex": "codex-medium",
-	anthropic: "claude-opus",
+	// Newest Anthropic flagship, and cheaper per token than the Opus 5 preset it
+	// replaces here. `claude-opus` stays selectable for the older cost/effort shape.
+	anthropic: "claude-opus-5-5",
 	"opencode-go": "opencodego",
 	zai: "glm-medium",
 	"kimi-code": "kimi-coding-plan-medium",

package/src/config/settings-schema.ts

@@ -2008,12 +2008,32 @@ export const SETTINGS_SCHEMA = {
 	// Typed decisions
 	"decisions.enabled": {
 		type: "boolean",
-		default: false,
+		default: true,
 		ui: {
 			tab: "context",
 			label: "Typed decisions",
 			description:
-				"Add a model-backed second stage to workflow routing. The keyword table already runs on every turn and costs nothing; this handles the phrasings it cannot express, which is most wording that is not a literal match. Costs one small model call, and only on turns the keyword table did not already answer. Any failure falls back to keyword-only behaviour, but a successful answer can also select a different workflow than the deep-interview ambiguity detector would have.",
+				"Model-backed second stage for workflow routing. The keyword table runs first on every turn and costs nothing; this handles the phrasings it cannot express, which is most wording that is not a literal match. Costs one small model call, only on turns the keyword table did not already answer, on the cheapest backend available: TypeSafe when a key is stored, else a local runtime that is running, else the small model of the provider you are chatting with. Any failure falls back to keyword-only behaviour, but a successful answer can also select a different workflow than the deep-interview ambiguity detector would have. Turn off to route on the keyword table alone.",
+		},
+	},
+
+	/**
+	 * Feed routing answers back into the deterministic keyword table.
+	 *
+	 * The hand-written table cannot be grown by hand for Korean — measured recall
+	 * was 0/9 — so it grows itself instead: a two-stem pattern that produced the
+	 * same routing answer on two distinct prompts, and was never contradicted, is
+	 * promoted and thereafter fires for free. One contradiction retracts it
+	 * permanently. Stems and hashes are stored; prompt text never is.
+	 */
+	"decisions.keywordLearning": {
+		type: "boolean",
+		default: true,
+		ui: {
+			tab: "context",
+			label: "Learn routing keywords",
+			description:
+				"Remember the phrasings the routing model resolves, so repeating one stops costing a model call. A pattern must give the same answer on two different prompts before it fires on its own, and a single disagreement removes it for good. Only word stems and hashes are written to disk, never your prompts. Turn off to keep the keyword table frozen at its built-in entries.",
 		},
 	},
 
@@ -2026,12 +2046,14 @@ export const SETTINGS_SCHEMA = {
 	 * mid-session invalidates the prompt cache, which on a long context costs
 	 * more than the cheaper tier saves.
 	 *
-	 * Needs `decisions.enabled` and at least two tiers configured. Without both
-	 * it never fires and the configured role models are used unchanged.
+	 * Needs `decisions.enabled` and at least two tiers configured. The tiers are
+	 * empty by default, so this is a no-op until the user sets them — which is why
+	 * it defaults on: shipping it off meant the routing code existed and never ran
+	 * even for users who had configured the tiers it needs.
 	 */
 	"task.modelRouting.enabled": {
 		type: "boolean",
-		default: false,
+		default: true,
 		ui: {
 			tab: "tasks",
 			label: "Route subagent models per task",

package/src/decisions/index.ts

@@ -43,8 +43,9 @@ export function createDecisionService(options: DecisionServiceOptions): Decision
 	 * probabilities, and it resolves `null` immediately when no key is stored, so users
 	 * who never added one pay nothing for it being in the list.
 	 *
-	 * The user's logged-in model is the fallback and the default experience: no extra
-	 * vendor, no extra key, works offline of TypeSafe entirely.
+	 * The user's logged-in provider is the fallback and the default experience: no extra
+	 * vendor, no extra key, works offline of TypeSafe entirely. Inside it, a live local
+	 * runtime is tried before a hosted small model; see `llm-backend.ts`.
 	 */
 	const backends = options.backends ?? [createTypeSafeDecisionBackend(options), createLlmDecisionBackend(options)];
 
@@ -53,6 +54,11 @@ export function createDecisionService(options: DecisionServiceOptions): Decision
 		async decide(request: DecisionRequest): Promise<DecisionResult | null> {
 			if (!enabled || backends.length === 0) return null;
 			for (const backend of backends) {
+				// A listener added to a signal that already fired never runs, and the
+				// backends only read their own derived signal. Measured: a caller that
+				// aborted during the previous backend's await got a full attempt budget of
+				// silence from the next one instead of an immediate null.
+				if (request.signal?.aborted) return null;
 				const controller = new AbortController();
 				const abortOnCallerSignal = () => controller.abort();
 				request.signal?.addEventListener("abort", abortOnCallerSignal, { once: true });