@dimi-agent/cli 0.5.5 → 0.6.0

package/dist/main.mjs

@@ -121085,6 +121085,56 @@ var init_toolPolicyService = __esmMin((() => {
 	registerScopedService(2, IAgentToolPolicyService, AgentToolPolicyService, 0, "toolPolicy");
 }));
 //#endregion
+//#region ../../packages/agent-core-v2/src/agent/loop/contextSize.ts
+/**
+* Scale a token count by `percent` of the model's default window, keeping
+* at least `CONTEXT_SIZE_FLOOR_TOKENS` — but never more than the original
+* window, so models already below the floor (and `percent` at or above
+* 100, or unset) pass through unchanged.
+*/
+function scaleContextTokens(tokens, percent) {
+	if (!Number.isFinite(tokens) || tokens <= 0) return tokens;
+	if (percent === void 0 || percent <= 0 || percent >= 100) return tokens;
+	const scaled = Math.floor(tokens * percent / 100);
+	return Math.max(scaled, Math.min(tokens, CONTEXT_SIZE_FLOOR_TOKENS));
+}
+/**
+* Apply the configured context-size percentage to a model capability.
+* Returns the original object unchanged when nothing changes, so
+* `UNKNOWN_CAPABILITY` identity and other callers relying on object
+* equality keep working.
+*/
+function scaleModelCapabilityContext(capability, percent) {
+	if (percent === void 0 || percent <= 0 || percent >= 100) return capability;
+	const maxContextTokens = scaleContextTokens(capability.max_context_tokens, percent);
+	const maxInputTokens = capability.max_input_tokens === void 0 ? void 0 : scaleContextTokens(capability.max_input_tokens, percent);
+	if (maxContextTokens === capability.max_context_tokens && maxInputTokens === capability.max_input_tokens) return capability;
+	return {
+		...capability,
+		max_context_tokens: maxContextTokens,
+		max_input_tokens: maxInputTokens
+	};
+}
+/**
+* Offered percentage levels for a model window: `100`, `95`, … descending
+* in `CONTEXT_SIZE_STEP_PERCENT` steps while the scaled size stays at or
+* above `CONTEXT_SIZE_FLOOR_TOKENS`. Empty when the window is below the
+* floor — the model's context size is not adjustable.
+*/
+function contextSizePercentOptions(contextWindow) {
+	if (!Number.isFinite(contextWindow) || contextWindow <= 0) return [];
+	const options = [];
+	for (let percent = 100; percent >= 5; percent -= 5) {
+		if (Math.floor(contextWindow * percent / 100) < 2e5) break;
+		options.push(percent);
+	}
+	return options;
+}
+var CONTEXT_SIZE_FLOOR_TOKENS;
+var init_contextSize$1 = __esmMin((() => {
+	CONTEXT_SIZE_FLOOR_TOKENS = 2e5;
+}));
+//#endregion
 //#region ../../packages/agent-core-v2/src/agent/loop/configSection.ts
 function parseNonNegativeInt(raw) {
 	const value = raw.trim();
@@ -121098,6 +121148,7 @@ var init_configSection$5 = __esmMin((() => {
 	init_config$5();
 	init_configSectionContributions();
 	init_toml();
+	init_contextSize$1();
 	LOOP_CONTROL_SECTION = "loopControl";
 	LOOP_MAX_STEPS_PER_TURN_ENV = "DIMI_LOOP_MAX_STEPS_PER_TURN";
 	LOOP_MAX_RETRIES_PER_STEP_ENV = "DIMI_LOOP_MAX_RETRIES_PER_STEP";
@@ -121106,7 +121157,13 @@ var init_configSection$5 = __esmMin((() => {
 		maxRetriesPerStep: number$2().int().min(0).optional(),
 		maxRalphIterations: number$2().int().min(-1).optional(),
 		reservedContextSize: number$2().int().min(0).optional(),
-		compactionTriggerRatio: number$2().min(.5).max(.99).optional()
+		compactionTriggerRatio: number$2().min(.5).max(.99).optional(),
+		/**
+		* Effective context window as a percentage of the model's default, in
+		* 5% steps, never below `CONTEXT_SIZE_FLOOR_TOKENS`. See
+		* `agent/loop/contextSize.ts` for the scaling semantics.
+		*/
+		contextSizePercent: number$2().int().min(5).max(100).multipleOf(5).optional()
 	});
 	loopControlEnvBindings = envBindings(LoopControlSchema, {
 		maxStepsPerTurn: {
@@ -211855,6 +211912,8 @@ var init_profileService = __esmMin((() => {
 	init_errors$11();
 	init_bootstrap();
 	init_config$5();
+	init_configSection$5();
+	init_contextSize$1();
 	init_hostEnvironment();
 	init_hostFileSystem();
 	init_sessionContext();
@@ -211942,6 +212001,7 @@ var init_profileService = __esmMin((() => {
 					this.publishToolPatternWarnings();
 					this.refreshSystemPrompt();
 				}
+				if (domain === "loopControl") this.emitStatusUpdated();
 			}));
 			this._register(this.skillCatalog.onDidChange((sourceId) => {
 				if (sourceId === "plugin") this.refreshSystemPrompt();
@@ -212139,7 +212199,7 @@ var init_profileService = __esmMin((() => {
 			return {
 				cwd: this.cwd,
 				modelAlias: this.modelAlias,
-				modelCapabilities: model === void 0 ? UNKNOWN_CAPABILITY : modelCapabilities(model),
+				modelCapabilities: this.capabilitiesFor(model),
 				profileName: this.profileName,
 				thinkingLevel: this.thinkingLevel,
 				systemPrompt: this.systemPrompt,
@@ -212154,10 +212214,10 @@ var init_profileService = __esmMin((() => {
 		resolveModelContext() {
 			const modelAlias = this.model;
 			const model = this.modelCatalog.get(modelAlias);
-			const loopControl = this.config.get("loopControl");
+			const loopControl = this.config.get(LOOP_CONTROL_SECTION);
 			return {
 				modelAlias,
-				modelCapabilities: modelCapabilities(model),
+				modelCapabilities: this.capabilitiesFor(model),
 				maxOutputSize: model.maxTokens,
 				alwaysThinking: modelAlwaysThinking(model) || void 0,
 				thinkingLevel: this.resolveThinkingState(model).effective,
@@ -212182,8 +212242,15 @@ var init_profileService = __esmMin((() => {
 			};
 		}
 		getModelCapabilities() {
-			const model = this.tryResolveRawModel();
-			return model === void 0 ? UNKNOWN_CAPABILITY : modelCapabilities(model);
+			return this.capabilitiesFor(this.tryResolveRawModel());
+		}
+		/**
+		* Model capabilities with the user's `[loop_control] context_size_percent`
+		* applied to the token limits. The percentage is read on every call, so
+		* config changes take effect on the next request without a restart.
+		*/
+		capabilitiesFor(model) {
+			return scaleModelCapabilityContext(model === void 0 ? UNKNOWN_CAPABILITY : modelCapabilities(model), this.config.get(LOOP_CONTROL_SECTION)?.contextSizePercent);
 		}
 		getMaxOutputSize() {
 			return this.tryResolveRawModel()?.maxTokens;
@@ -217678,6 +217745,7 @@ var init_src$6 = __esmMin((() => {
 	init_retry();
 	init_configSection$5();
 	init_configSection$5();
+	init_contextSize$1();
 	init_loop();
 	init_loopService();
 	init_loopContinuation();
@@ -309726,6 +309794,134 @@ var init_busy_input_mode_selector = __esmMin((() => {
 	};
 }));
 //#endregion
+//#region src/utils/usage/usage-format.ts
+function usageNumber$1(value) {
+	return typeof value === "number" && Number.isFinite(value) && value > 0 ? value : 0;
+}
+/**
+* Total prompt tokens for a single generation: non-cache input + cache
+* read + cache write/creation. Matches the denominator pi uses for CH%.
+*/
+function promptTokenTotal(usage) {
+	if (usage === null || usage === void 0) return 0;
+	return usageNumber$1(usage.inputOther) + usageNumber$1(usage.inputCacheRead) + usageNumber$1(usage.inputCacheCreation);
+}
+/**
+* Latest-step prompt-cache hit rate as a percentage in [0, 100].
+* Returns `undefined` when there is no prompt traffic, or when the
+* provider reported neither cache reads nor cache writes (no cache
+* signal — same gate as pi's footer `CH` badge).
+*/
+function cacheHitRatePercent(usage) {
+	if (usage === null || usage === void 0) return void 0;
+	const cacheRead = usageNumber$1(usage.inputCacheRead);
+	const cacheWrite = usageNumber$1(usage.inputCacheCreation);
+	if (cacheRead === 0 && cacheWrite === 0) return void 0;
+	const prompt = promptTokenTotal(usage);
+	if (prompt <= 0) return void 0;
+	return Math.min(100, Math.max(0, cacheRead / prompt * 100));
+}
+/** Compact footer/report label, e.g. `CH85.0%`. */
+function formatCacheHitRate(rate) {
+	if (!Number.isFinite(rate)) return "CH?%";
+	return `CH${rate.toFixed(1)}%`;
+}
+/**
+* Format a token count in 1024-based units: context sizes are powers of
+* two, so 262144 reads as "256k", not "262.1k". k values at or above
+* 100 are rounded to whole numbers ("977k").
+*/
+function formatTokenCount(n) {
+	if (!Number.isFinite(n) || n < 0) return "0";
+	if (n >= 1024 * 1024) return `${trimDecimal(n / (1024 * 1024))}M`;
+	if (n >= 1024) {
+		const k = n / 1024;
+		return `${k >= 100 ? Math.round(k) : trimDecimal(k)}k`;
+	}
+	return String(n);
+}
+/** One decimal place, dropping a redundant ".0" ("1.0" → "1", "1.5" stays). */
+function trimDecimal(v) {
+	const s = v.toFixed(1);
+	return s.endsWith(".0") ? s.slice(0, -2) : s;
+}
+/**
+* Format a token count in 1000-based units ("1M", "950k", "200k"). Unlike
+* `formatTokenCount` (1024-based, the footer/usage convention), this keeps
+* exact product numbers readable: the context-size picker's floor is a
+* literal 200k, and 5% steps of a 1M window land on round values.
+*/
+function formatDecimalTokenCount(n) {
+	if (!Number.isFinite(n) || n < 0) return "0";
+	if (n >= 1e6) return `${trimDecimal(n / 1e6)}M`;
+	if (n >= 1e3) return `${trimDecimal(n / 1e3)}k`;
+	return String(n);
+}
+/**
+* Usage as a whole-number percentage of `max`, ceiled so any non-zero
+* usage shows at least 1%, clamped to [0, 100]. A non-positive or
+* non-finite `max` reports 0.
+*/
+function usagePercent(used, max) {
+	if (!Number.isFinite(max) || max <= 0) return 0;
+	return Math.min(100, Math.max(0, Math.ceil(used / max * 100)));
+}
+/** `usagePercent` for callers that only know the ratio (NaN-safe). */
+function usagePercentFromRatio(ratio) {
+	return Math.min(100, Math.max(0, Math.ceil(safeUsageRatio(ratio) * 100)));
+}
+/**
+* Build a `[███░░░░░░░]` style bar. Returns a plain-ASCII string with
+* `filled`/`empty` glyphs — colouring is the caller's responsibility.
+*/
+function renderProgressBar(ratio, width = 20, filled = "█", empty = "░") {
+	const clamped = safeUsageRatio(ratio);
+	const filledCount = Math.round(clamped * width);
+	return filled.repeat(filledCount) + empty.repeat(Math.max(0, width - filledCount));
+}
+function safeUsageRatio(ratio) {
+	return Number.isFinite(ratio) ? Math.max(0, Math.min(ratio, 1)) : 0;
+}
+/**
+* Map a usage ratio to a semantic colour token — the `/usage` renderer
+* translates these into palette hex values.
+*/
+function ratioSeverity(ratio) {
+	if (ratio >= .85) return "danger";
+	if (ratio >= .5) return "warn";
+	return "ok";
+}
+var init_usage_format = __esmMin((() => {}));
+//#endregion
+//#region src/tui/components/dialogs/context-size-selector.ts
+function buildOptions(contextWindow, percentOptions) {
+	return percentOptions.map((percent) => ({
+		value: String(percent),
+		label: percent === 100 ? "100% (default)" : `${percent}%`,
+		description: `${formatDecimalTokenCount(Math.floor(contextWindow * percent / 100))} tokens`
+	}));
+}
+var ContextSizeSelectorComponent;
+var init_context_size_selector = __esmMin((() => {
+	init_src$6();
+	init_usage_format();
+	init_choice_picker();
+	ContextSizeSelectorComponent = class extends ChoicePickerComponent {
+		constructor(opts) {
+			super({
+				title: "Context size",
+				hint: `Model window ${formatDecimalTokenCount(opts.contextWindow)} · floor ${formatDecimalTokenCount(CONTEXT_SIZE_FLOOR_TOKENS)}`,
+				options: buildOptions(opts.contextWindow, opts.percentOptions),
+				currentValue: String(opts.currentPercent),
+				onSelect: (value) => {
+					opts.onSelect(Number(value));
+				},
+				onCancel: opts.onCancel
+			});
+		}
+	};
+}));
+//#endregion
 //#region src/tui/components/dialogs/editor-selector.ts
 var EDITOR_OPTIONS, EditorSelectorComponent;
 var init_editor_selector = __esmMin((() => {
@@ -310186,7 +310382,7 @@ var init_permission_selector = __esmMin((() => {
 //#endregion
 //#region src/tui/components/dialogs/settings-selector.ts
 function isSettingsSelection(value) {
-	return value === "model" || value === "theme" || value === "editor" || value === "permission" || value === "busy-input" || value === "experiments" || value === "upgrade" || value === "usage";
+	return value === "model" || value === "theme" || value === "editor" || value === "permission" || value === "busy-input" || value === "context-size" || value === "experiments" || value === "upgrade" || value === "usage";
 }
 var SETTINGS_OPTIONS, SettingsSelectorComponent;
 var init_settings_selector = __esmMin((() => {
@@ -310217,6 +310413,11 @@ var init_settings_selector = __esmMin((() => {
 			label: "Busy input",
 			description: "Choose whether Enter queues or steers while the agent is working."
 		},
+		{
+			value: "context-size",
+			label: "Context size",
+			description: "Cap the conversation window as a percentage of the model default (min 200k)."
+		},
 		{
 			value: "experiments",
 			label: "Experiments",
@@ -310462,93 +310663,6 @@ var init_mcp_status_panel = __esmMin((() => {
 	];
 }));
 //#endregion
-//#region src/utils/usage/usage-format.ts
-function usageNumber$1(value) {
-	return typeof value === "number" && Number.isFinite(value) && value > 0 ? value : 0;
-}
-/**
-* Total prompt tokens for a single generation: non-cache input + cache
-* read + cache write/creation. Matches the denominator pi uses for CH%.
-*/
-function promptTokenTotal(usage) {
-	if (usage === null || usage === void 0) return 0;
-	return usageNumber$1(usage.inputOther) + usageNumber$1(usage.inputCacheRead) + usageNumber$1(usage.inputCacheCreation);
-}
-/**
-* Latest-step prompt-cache hit rate as a percentage in [0, 100].
-* Returns `undefined` when there is no prompt traffic, or when the
-* provider reported neither cache reads nor cache writes (no cache
-* signal — same gate as pi's footer `CH` badge).
-*/
-function cacheHitRatePercent(usage) {
-	if (usage === null || usage === void 0) return void 0;
-	const cacheRead = usageNumber$1(usage.inputCacheRead);
-	const cacheWrite = usageNumber$1(usage.inputCacheCreation);
-	if (cacheRead === 0 && cacheWrite === 0) return void 0;
-	const prompt = promptTokenTotal(usage);
-	if (prompt <= 0) return void 0;
-	return Math.min(100, Math.max(0, cacheRead / prompt * 100));
-}
-/** Compact footer/report label, e.g. `CH85.0%`. */
-function formatCacheHitRate(rate) {
-	if (!Number.isFinite(rate)) return "CH?%";
-	return `CH${rate.toFixed(1)}%`;
-}
-/**
-* Format a token count in 1024-based units: context sizes are powers of
-* two, so 262144 reads as "256k", not "262.1k". k values at or above
-* 100 are rounded to whole numbers ("977k").
-*/
-function formatTokenCount(n) {
-	if (!Number.isFinite(n) || n < 0) return "0";
-	if (n >= 1024 * 1024) return `${trimDecimal(n / (1024 * 1024))}M`;
-	if (n >= 1024) {
-		const k = n / 1024;
-		return `${k >= 100 ? Math.round(k) : trimDecimal(k)}k`;
-	}
-	return String(n);
-}
-/** One decimal place, dropping a redundant ".0" ("1.0" → "1", "1.5" stays). */
-function trimDecimal(v) {
-	const s = v.toFixed(1);
-	return s.endsWith(".0") ? s.slice(0, -2) : s;
-}
-/**
-* Usage as a whole-number percentage of `max`, ceiled so any non-zero
-* usage shows at least 1%, clamped to [0, 100]. A non-positive or
-* non-finite `max` reports 0.
-*/
-function usagePercent(used, max) {
-	if (!Number.isFinite(max) || max <= 0) return 0;
-	return Math.min(100, Math.max(0, Math.ceil(used / max * 100)));
-}
-/** `usagePercent` for callers that only know the ratio (NaN-safe). */
-function usagePercentFromRatio(ratio) {
-	return Math.min(100, Math.max(0, Math.ceil(safeUsageRatio(ratio) * 100)));
-}
-/**
-* Build a `[███░░░░░░░]` style bar. Returns a plain-ASCII string with
-* `filled`/`empty` glyphs — colouring is the caller's responsibility.
-*/
-function renderProgressBar(ratio, width = 20, filled = "█", empty = "░") {
-	const clamped = safeUsageRatio(ratio);
-	const filledCount = Math.round(clamped * width);
-	return filled.repeat(filledCount) + empty.repeat(Math.max(0, width - filledCount));
-}
-function safeUsageRatio(ratio) {
-	return Number.isFinite(ratio) ? Math.max(0, Math.min(ratio, 1)) : 0;
-}
-/**
-* Map a usage ratio to a semantic colour token — the `/usage` renderer
-* translates these into palette hex values.
-*/
-function ratioSeverity(ratio) {
-	if (ratio >= .85) return "danger";
-	if (ratio >= .5) return "warn";
-	return "ok";
-}
-var init_usage_format = __esmMin((() => {}));
-//#endregion
 //#region src/tui/components/messages/usage-panel.ts
 function usageRowLabel(row) {
 	const window = row.window;
@@ -312313,6 +312427,62 @@ function showBusyInputModePicker(host) {
 		}
 	}));
 }
+/** `[loop_control]` from a `getConfig()` result (untyped domain access). */
+function loopControlFromConfig(config) {
+	return config["loopControl"];
+}
+/** The active model's default context window in tokens, or 0 when unknown. */
+function activeModelContextWindow(host) {
+	const alias = host.state.appState.model;
+	return (alias === void 0 ? void 0 : host.state.appState.availableModels[alias])?.maxContextSize ?? 0;
+}
+/**
+* Open the context-size picker: 5% steps from the model's default window
+* (100%) down while the scaled window stays at or above 200k. Models whose
+* window is already below 200k are not adjustable — a notice explains why
+* instead of opening a one-option picker.
+*/
+function showContextSizePicker(host) {
+	const contextWindow = activeModelContextWindow(host);
+	if (contextWindow <= 0) {
+		host.showNotice("Context size unavailable", "No model is active. Choose a model first.");
+		return;
+	}
+	const percentOptions = contextSizePercentOptions(contextWindow);
+	if (percentOptions.length <= 1) {
+		host.showNotice("Context size cannot be adjusted", `This model's window (${formatDecimalTokenCount(contextWindow)}) is already at or below the ${formatDecimalTokenCount(CONTEXT_SIZE_FLOOR_TOKENS)} floor.`);
+		return;
+	}
+	host.harness.getConfig().then((config) => {
+		const currentPercent = loopControlFromConfig(config)?.contextSizePercent ?? 100;
+		host.mountEditorReplacement(new ContextSizeSelectorComponent({
+			contextWindow,
+			percentOptions,
+			currentPercent,
+			onSelect: (percent) => {
+				host.restoreEditor();
+				applyContextSizeChoice(host, percent);
+			},
+			onCancel: () => {
+				host.restoreEditor();
+			}
+		}));
+	}, (error) => {
+		host.showError(`Failed to load context size: ${formatErrorMessage$2(error)}`);
+	});
+}
+async function applyContextSizeChoice(host, percent) {
+	const contextWindow = activeModelContextWindow(host);
+	try {
+		await host.harness.setConfig({ loopControl: { contextSizePercent: percent } });
+	} catch (error) {
+		host.showError(`Failed to save context size: ${formatErrorMessage$2(error)}`);
+		return;
+	}
+	const maxContextTokens = contextWindow > 0 ? scaleContextTokens(contextWindow, percent) : 0;
+	if (maxContextTokens > 0) host.setAppState({ maxContextTokens });
+	host.showNotice(`Context size: ${percent}%`, `Effective window ${formatDecimalTokenCount(maxContextTokens)} (default ${formatDecimalTokenCount(contextWindow)}).`);
+}
 async function showExperimentsPanel(host) {
 	let features;
 	try {
@@ -312457,6 +312627,9 @@ function handleSettingsSelection(host, value) {
 		case "busy-input":
 			showBusyInputModePicker(host);
 			return;
+		case "context-size":
+			showContextSizePicker(host);
+			return;
 		case "experiments":
 			showExperimentsPanel(host);
 			return;
@@ -312472,6 +312645,7 @@ var MODEL_PICKER_REFRESH_TIMEOUT_MS, MODEL_SWITCH_CACHE_WARNING, EFFORT_SWITCH_C
 var init_config = __esmMin((() => {
 	init_src$4();
 	init_busy_input_mode_selector();
+	init_context_size_selector();
 	init_editor_selector();
 	init_effort_selector();
 	init_experiments_selector();
@@ -312486,6 +312660,8 @@ var init_config = __esmMin((() => {
 	init_dimi_tui$1();
 	init_event_payload();
 	init_thinking_config();
+	init_src$6();
+	init_usage_format();
 	init_info();
 	init_experimental_flags();
 	MODEL_PICKER_REFRESH_TIMEOUT_MS = 2e3;

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@dimi-agent/cli",
-  "version": "0.5.5",
+  "version": "0.6.0",
   "description": "The Starting Point for Next-Gen Agents",
   "keywords": [
     "agent",
@@ -60,14 +60,14 @@
     "tsx": "^4.21.0",
     "yazl": "^3.3.1",
     "zod": "^4.3.6",
+    "@dimi-agent/agent-core-v2": "^0.1.0",
     "@dimi-agent/dimi-oauth": "^0.1.0",
     "@dimi-agent/dimi-sdk": "^0.2.0",
     "@dimi-agent/dimi-telemetry": "^0.1.0",
-    "@dimi-agent/agent-core-v2": "^0.1.0",
     "@dimi-agent/dimi-web": "^0.1.0",
-    "@dimi-agent/pi-tui": "^0.1.0",
+    "@dimi-agent/kap-server": "^0.1.0",
     "@dimi-agent/remote": "^0.1.0",
-    "@dimi-agent/kap-server": "^0.1.0"
+    "@dimi-agent/pi-tui": "^0.1.0"
   },
   "optionalDependencies": {
     "@mariozechner/clipboard": "^0.3.9",