muonroi-cli 1.8.3 → 1.8.5

package/dist/src/scaffold/point-to-existing.d.ts

@@ -5,6 +5,7 @@
  * Designed for testability: callers inject detectVerifyRecipe so unit tests
  * never hit the real orchestrator.
  */
+import type { VerifyRecipe } from "../types/index.js";
 import { detectBBFramework } from "./init-new.js";
 export interface PointToExistingOptions {
     /** Raw path string entered by the user. May be relative or absolute. */
@@ -26,6 +27,26 @@ export interface PointToExistingResult {
     targetFramework?: "muonroi-building-block" | string;
 }
 export { detectBBFramework };
+/**
+ * Deterministic, filesystem-only verify-recipe detection for an existing project
+ * the user pointed to via the CB-3 recovery card.
+ *
+ * WHY NOT the orchestrator's detectVerifyRecipe: that path runs an LLM
+ * verify-detect sub-agent against the orchestrator's OWN (session) cwd via
+ * `this.bash.getCwd()` — it cannot inspect an arbitrary pointed-to directory, so
+ * wiring it here was deferred and the handler stubbed to `null` (point-to-existing
+ * therefore ALWAYS reported "no recipe", making the recovery option inert).
+ * `inferVerifyProjectProfile` reads the target dir's manifest/build files
+ * directly (package.json test/check/lint, Makefile, pyproject, go.mod, cargo,
+ * maven/gradle, *.csproj) — the correct detector for adopting an existing repo.
+ *
+ * Returns the recipe only when it is actually RUNNABLE (has a test/build/start
+ * command). The bare "unknown" fallback that `inferVerifyProjectProfile` emits
+ * for a directory with no recognizable project files carries empty commands →
+ * we return null so point-to-existing correctly reports `no_recipe` instead of
+ * falsely adopting an empty directory.
+ */
+export declare function detectExistingProjectRecipe(cwd: string): VerifyRecipe | null;
 /**
  * Resolve and validate a user-provided path for sprint re-entry.
  *

package/dist/src/scaffold/point-to-existing.js

@@ -7,8 +7,33 @@
  */
 import { realpathSync, statSync } from "node:fs";
 import * as path from "node:path";
+import { inferVerifyProjectProfile } from "../verify/recipes.js";
 import { detectBBFramework } from "./init-new.js";
 export { detectBBFramework };
+/**
+ * Deterministic, filesystem-only verify-recipe detection for an existing project
+ * the user pointed to via the CB-3 recovery card.
+ *
+ * WHY NOT the orchestrator's detectVerifyRecipe: that path runs an LLM
+ * verify-detect sub-agent against the orchestrator's OWN (session) cwd via
+ * `this.bash.getCwd()` — it cannot inspect an arbitrary pointed-to directory, so
+ * wiring it here was deferred and the handler stubbed to `null` (point-to-existing
+ * therefore ALWAYS reported "no recipe", making the recovery option inert).
+ * `inferVerifyProjectProfile` reads the target dir's manifest/build files
+ * directly (package.json test/check/lint, Makefile, pyproject, go.mod, cargo,
+ * maven/gradle, *.csproj) — the correct detector for adopting an existing repo.
+ *
+ * Returns the recipe only when it is actually RUNNABLE (has a test/build/start
+ * command). The bare "unknown" fallback that `inferVerifyProjectProfile` emits
+ * for a directory with no recognizable project files carries empty commands →
+ * we return null so point-to-existing correctly reports `no_recipe` instead of
+ * falsely adopting an empty directory.
+ */
+export function detectExistingProjectRecipe(cwd) {
+    const { recipe } = inferVerifyProjectProfile(cwd);
+    const runnable = recipe.testCommands.length > 0 || recipe.buildCommands.length > 0 || Boolean(recipe.startCommand);
+    return runnable ? recipe : null;
+}
 // ---------------------------------------------------------------------------
 // Implementation
 // ---------------------------------------------------------------------------

package/dist/src/self-qa/agentic-loop.js

@@ -65,25 +65,25 @@ export function createMockBrain(script) {
 // ---------------------------------------------------------------------------
 // LLM brain — uses the configured provider stack
 // ---------------------------------------------------------------------------
-const DEFAULT_SYSTEM_PROMPT = `You are a QA agent driving a TUI via an agent-harness.
-Your job: given a free-form goal and the current TUI state (semantic tree + recent events),
-decide the SINGLE next action.
-
-Reply with a strict JSON object — no prose, no markdown fences:
-
-  {"action":"type","text":"...","reason":"short"}
-  {"action":"press","key":"Enter|Escape|Down|Up|Tab","reason":"short"}
-  {"action":"wait_for","idle":true,"timeoutMs":5000,"reason":"..."}
-  {"action":"wait_for","selector":"id=askcard","timeoutMs":5000,"reason":"..."}
-  {"action":"wait_for","event":"askcard-open","timeoutMs":5000,"reason":"..."}
-  {"action":"done","verdict":"pass","reason":"goal achieved"}
-  {"action":"done","verdict":"fail","reason":"goal cannot be achieved"}
-
-Rules:
-- One action per turn. Do not batch.
-- Prefer wait_for after type/press to let the TUI settle.
-- If the goal looks satisfied, return done(pass).
-- If an error toast appears or the TUI is stuck, return done(fail).
+const DEFAULT_SYSTEM_PROMPT = `You are a QA agent driving a TUI via an agent-harness.
+Your job: given a free-form goal and the current TUI state (semantic tree + recent events),
+decide the SINGLE next action.
+
+Reply with a strict JSON object — no prose, no markdown fences:
+
+  {"action":"type","text":"...","reason":"short"}
+  {"action":"press","key":"Enter|Escape|Down|Up|Tab","reason":"short"}
+  {"action":"wait_for","idle":true,"timeoutMs":5000,"reason":"..."}
+  {"action":"wait_for","selector":"id=askcard","timeoutMs":5000,"reason":"..."}
+  {"action":"wait_for","event":"askcard-open","timeoutMs":5000,"reason":"..."}
+  {"action":"done","verdict":"pass","reason":"goal achieved"}
+  {"action":"done","verdict":"fail","reason":"goal cannot be achieved"}
+
+Rules:
+- One action per turn. Do not batch.
+- Prefer wait_for after type/press to let the TUI settle.
+- If the goal looks satisfied, return done(pass).
+- If an error toast appears or the TUI is stuck, return done(fail).
 - Stay within ${"${maxTurns}"} turns. Be concise.`;
 /**
  * Pull a usable text body out of an AI SDK generateText result.
@@ -148,8 +148,8 @@ export async function createLLMBrain(opts) {
     const retryOnEmpty = opts.retryOnEmpty !== false;
     // @ai-sdk/openai-compatible's chat schema exposes `strictJsonSchema: false`
     // for providers that don't honour OpenAI's strict JSON schema mode.
-    // DeepSeek (and DeepSeek-compatible providers like SiliconFlow) only
-    // support `response_format: {type: "json_object"}` — they reject the full
+    // DeepSeek only supports `response_format: {type: "json_object"}` — it
+    // rejects the full
     // schema response_format. Setting strictJsonSchema=false makes AI SDK fall
     // back to lenient json_object mode silently (no "responseFormat not
     // supported" warning) while still steering the model toward JSON output.
@@ -161,7 +161,7 @@ export async function createLLMBrain(opts) {
     const mergedProviderOptions = {
         ...(runtime.providerOptions ?? {}),
     };
-    if (provider === "deepseek" || provider === "siliconflow") {
+    if (provider === "deepseek") {
         mergedProviderOptions[provider] = {
             ...(mergedProviderOptions[provider] ?? {}),
             strictJsonSchema: false,

package/dist/src/{ui/state → state}/active-run.d.ts

@@ -21,6 +21,25 @@ export declare const activeRunStore: {
     clearActiveRun: () => void;
     subscribe: (fn: Listener) => (() => void);
 };
+/**
+ * Single canonical writer for "which run is the workspace focused on".
+ * Writes the durable pointer (`state.md` `Active Run`) AND the in-memory cache
+ * atomically, so disk and StatusBar/reporter never drift (F8 follow-up).
+ *
+ * The disk write is idempotent with `ensureRunScoped`'s earlier write for the
+ * same runId — routing the focus moment through here just guarantees they agree.
+ */
+export declare function setWorkspaceFocus(flowDir: string, focus: {
+    runId: string;
+    productSlug: string;
+    reason?: string;
+}): Promise<void>;
+/**
+ * Clear the workspace focus cache when a run ends (shipped / halted).
+ * Cache-only by design: the durable `Active Run` pointer is left intact as a
+ * historical marker so `/ideal phases` still resolves the last run post-ship.
+ */
+export declare function clearWorkspaceFocus(): void;
 /** Test helper: reset to initial state. */
 export declare function __resetActiveRunStoreForTests(): void;
 export {};

package/dist/src/{ui/state → state}/active-run.js

@@ -37,6 +37,27 @@ function makeActiveRunStore() {
     };
 }
 export const activeRunStore = makeActiveRunStore();
+/**
+ * Single canonical writer for "which run is the workspace focused on".
+ * Writes the durable pointer (`state.md` `Active Run`) AND the in-memory cache
+ * atomically, so disk and StatusBar/reporter never drift (F8 follow-up).
+ *
+ * The disk write is idempotent with `ensureRunScoped`'s earlier write for the
+ * same runId — routing the focus moment through here just guarantees they agree.
+ */
+export async function setWorkspaceFocus(flowDir, focus) {
+    const { setActiveRunId } = await import("../flow/run-manager.js");
+    await setActiveRunId(flowDir, focus.runId);
+    activeRunStore.setActiveRun(focus.runId, flowDir, focus.productSlug);
+}
+/**
+ * Clear the workspace focus cache when a run ends (shipped / halted).
+ * Cache-only by design: the durable `Active Run` pointer is left intact as a
+ * historical marker so `/ideal phases` still resolves the last run post-ship.
+ */
+export function clearWorkspaceFocus() {
+    activeRunStore.clearActiveRun();
+}
 /** Test helper: reset to initial state. */
 export function __resetActiveRunStoreForTests() {
     activeRunStore.clearActiveRun();

package/dist/src/{ui/status-bar/store.d.ts → state/status-bar-store.d.ts}

@@ -1,9 +1,12 @@
 /**
- * src/ui/status-bar/store.ts
+ * src/state/status-bar-store.ts
  *
  * Status bar store: subscribable atom holding model/provider/tier/tokens/USD/degraded.
  * wireStatusBar() connects to routerStore (Plan 03), subscribeThresholds (Plan 04),
  * and subscribeDowngrade (Plan 05).
+ *
+ * Presentation-agnostic state — lives outside src/ui so the headless core
+ * (orchestrator/tool-engine/stream-runner) can read it without importing ui/.
  */
 export interface SprintProgressSegment {
     /** Active sprint index (1-based). */

package/dist/src/{ui/status-bar/store.js → state/status-bar-store.js}

@@ -1,15 +1,18 @@
 /**
- * src/ui/status-bar/store.ts
+ * src/state/status-bar-store.ts
  *
  * Status bar store: subscribable atom holding model/provider/tier/tokens/USD/degraded.
  * wireStatusBar() connects to routerStore (Plan 03), subscribeThresholds (Plan 04),
  * and subscribeDowngrade (Plan 05).
+ *
+ * Presentation-agnostic state — lives outside src/ui so the headless core
+ * (orchestrator/tool-engine/stream-runner) can read it without importing ui/.
  */
-import { getCircuitState } from "../../ee/client.js";
-import { routerStore } from "../../router/store.js";
-import { subscribeDowngrade } from "../../usage/downgrade.js";
-import { subscribeThresholds } from "../../usage/thresholds.js";
-import { activeRunStore } from "../state/active-run.js";
+import { getCircuitState } from "../ee/client.js";
+import { routerStore } from "../router/store.js";
+import { subscribeDowngrade } from "../usage/downgrade.js";
+import { subscribeThresholds } from "../usage/thresholds.js";
+import { activeRunStore } from "./active-run.js";
 function makeStore() {
     let state = {
         provider: "",
@@ -141,8 +144,8 @@ export function wireStatusBar() {
         }
         try {
             // Lazy import to avoid circular dependency at module load time.
-            const { readBacklog } = await import("../../product-loop/backlog-store.js");
-            const { readSprintPlan } = await import("../../product-loop/sprint-store.js");
+            const { readBacklog } = await import("../product-loop/backlog-store.js");
+            const { readSprintPlan } = await import("../product-loop/sprint-store.js");
             const [backlog, sprintPlan] = await Promise.all([
                 readBacklog(run.flowDir, run.runId).catch(() => null),
                 readSprintPlan(run.flowDir, run.runId).catch(() => null),
@@ -276,4 +279,4 @@ export function __resetStatusBarStoreForTests() {
     });
     wired = false;
 }
-//# sourceMappingURL=store.js.map
+//# sourceMappingURL=status-bar-store.js.map

package/dist/src/state/turn-trace.d.ts

@@ -0,0 +1,43 @@
+/**
+ * src/state/turn-trace.ts
+ *
+ * Pipeline debug-trace state — the toggle flag + recorded per-turn traces.
+ *
+ * Presentation-agnostic: the headless core (tool-engine, message-processor)
+ * records traces here without importing src/ui. The /debug slash command
+ * (src/ui/slash/debug.ts) owns the human formatting and toggling on top of
+ * this state.
+ */
+export interface PipelineStep {
+    name: string;
+    duration_ms: number;
+    input_summary: string;
+    output_summary: string;
+    tokens_saved?: number;
+}
+export interface TurnTrace {
+    turn_id: number;
+    timestamp: number;
+    raw_prompt: string;
+    steps: PipelineStep[];
+    model_requested: string;
+    model_used: string;
+    routed: boolean;
+    input_tokens: number;
+    output_tokens: number;
+    cache_read_tokens: number;
+    cost_usd: number;
+    estimated_savings: SavingsEstimate;
+}
+export interface SavingsEstimate {
+    pil_tokens_saved: number;
+    cache_tokens_saved: number;
+    router_cost_saved_usd: number;
+    total_tokens_saved: number;
+    total_cost_saved_usd: number;
+}
+export declare function isDebugEnabled(): boolean;
+export declare function setDebugEnabled(value: boolean): void;
+export declare function recordTurnTrace(trace: TurnTrace): void;
+export declare function getLastTrace(): TurnTrace | null;
+export declare function getAllTraces(): readonly TurnTrace[];

package/dist/src/state/turn-trace.js

@@ -0,0 +1,32 @@
+/**
+ * src/state/turn-trace.ts
+ *
+ * Pipeline debug-trace state — the toggle flag + recorded per-turn traces.
+ *
+ * Presentation-agnostic: the headless core (tool-engine, message-processor)
+ * records traces here without importing src/ui. The /debug slash command
+ * (src/ui/slash/debug.ts) owns the human formatting and toggling on top of
+ * this state.
+ */
+let debugEnabled = false;
+const turnTraces = [];
+export function isDebugEnabled() {
+    return debugEnabled;
+}
+export function setDebugEnabled(value) {
+    debugEnabled = value;
+}
+export function recordTurnTrace(trace) {
+    if (!debugEnabled)
+        return;
+    turnTraces.push(trace);
+    if (turnTraces.length > 50)
+        turnTraces.shift();
+}
+export function getLastTrace() {
+    return turnTraces[turnTraces.length - 1] ?? null;
+}
+export function getAllTraces() {
+    return turnTraces;
+}
+//# sourceMappingURL=turn-trace.js.map

package/dist/src/storage/db.js

@@ -9,7 +9,8 @@ let db = null;
 // async would be a much larger refactor.
 const requireSync = createRequire(import.meta.url);
 export function getDatabasePath() {
-    const dir = path.join(os.homedir(), ".muonroi-cli");
+    const home = process.env.HOME ?? process.env.USERPROFILE ?? os.homedir();
+    const dir = path.join(home, ".muonroi-cli");
     fs.mkdirSync(dir, { recursive: true, mode: 0o700 });
     return path.join(dir, "muonroi.db");
 }

package/dist/src/storage/index.d.ts

@@ -3,7 +3,7 @@ export { loadConfig, type MuonroiConfig } from "./config.js";
 export { getDatabasePath } from "./db.js";
 export { type InteractionEventType, logInteraction } from "./interaction-log.js";
 export { SessionStore } from "./sessions.js";
-export { appendCompaction, appendMessages, appendSystemMessage, buildChatEntries, getLastTodoWriteArgs, getNextMessageSequence, getSessionChain, loadLatestCompaction, loadRawTranscript, loadTranscript, loadTranscriptState, markMessageCompleted, markMessageErrored, markToolCallErrored, persistMessageWriteAhead, persistToolCallWriteAhead, revertLatestCompaction, } from "./transcript.js";
+export { appendCompaction, appendMessages, appendSystemMessage, buildChatEntries, getLastApprovedPlan, getLastTodoWriteArgs, getNextMessageSequence, getSessionChain, getSessionChainInfos, loadLatestCompaction, loadRawTranscript, loadSessionChainTranscriptState, loadTranscript, loadTranscriptState, markMessageCompleted, markMessageErrored, markToolCallErrored, persistApprovedPlan, persistMessageWriteAhead, persistToolCallWriteAhead, revertLatestCompaction, type SessionChainNode, } from "./transcript.js";
 export { buildEffectiveTranscript, type LoadedTranscriptState, type PersistedCompaction } from "./transcript-view.js";
 export { logUIInteraction, type UIInteractionSubtype } from "./ui-interaction-log.js";
 export { getSessionTotalTokens, listSessionUsage, recordUsageEvent, type TokenUsageLike } from "./usage.js";

package/dist/src/storage/index.js

@@ -4,7 +4,7 @@ export { loadConfig } from "./config.js";
 export { getDatabasePath } from "./db.js";
 export { logInteraction } from "./interaction-log.js";
 export { SessionStore } from "./sessions.js";
-export { appendCompaction, appendMessages, appendSystemMessage, buildChatEntries, getLastTodoWriteArgs, getNextMessageSequence, getSessionChain, loadLatestCompaction, loadRawTranscript, loadTranscript, loadTranscriptState, markMessageCompleted, markMessageErrored, markToolCallErrored, persistMessageWriteAhead, persistToolCallWriteAhead, revertLatestCompaction, } from "./transcript.js";
+export { appendCompaction, appendMessages, appendSystemMessage, buildChatEntries, getLastApprovedPlan, getLastTodoWriteArgs, getNextMessageSequence, getSessionChain, getSessionChainInfos, loadLatestCompaction, loadRawTranscript, loadSessionChainTranscriptState, loadTranscript, loadTranscriptState, markMessageCompleted, markMessageErrored, markToolCallErrored, persistApprovedPlan, persistMessageWriteAhead, persistToolCallWriteAhead, revertLatestCompaction, } from "./transcript.js";
 export { buildEffectiveTranscript } from "./transcript-view.js";
 export { logUIInteraction } from "./ui-interaction-log.js";
 export { getSessionTotalTokens, listSessionUsage, recordUsageEvent } from "./usage.js";

package/dist/src/storage/interaction-log.d.ts

@@ -4,7 +4,7 @@
  * Fire-and-forget interaction logging for detailed user-agent event tracking.
  * All calls are fail-open — logging never breaks the main flow.
  */
-export type InteractionEventType = "user_message" | "agent_response" | "tool_call" | "tool_result" | "compaction" | "routing" | "pil" | "error" | "model_switch" | "council" | "ee_intercept" | "ee_judge" | "ee_injection" | "ui_interaction" | "stream_retry" | "f6_synthesis" | "grounding_flag" | "stall_rescue" | "stream_start" | "text_tool_resteer" | "session_experience";
+export type InteractionEventType = "user_message" | "agent_response" | "tool_call" | "tool_result" | "compaction" | "routing" | "pil" | "error" | "model_switch" | "council" | "ee_intercept" | "ee_judge" | "ee_injection" | "ui_interaction" | "stream_retry" | "f6_synthesis" | "grounding_flag" | "stall_rescue" | "stream_start" | "text_tool_resteer" | "turn_tool_load" | "session_experience";
 export interface EEInjectionRow {
     session_id: string;
     event_subtype: string | null;

package/dist/src/storage/interaction-log.js

@@ -55,10 +55,10 @@ export function selectEEInjectionsForRun(runId, limit = 20) {
     try {
         const db = getDatabase();
         return db
-            .prepare(`SELECT session_id, event_subtype, duration_ms, metadata_json, created_at
-         FROM interaction_logs
-         WHERE session_id = ? AND event_type = 'ee_injection'
-         ORDER BY created_at DESC
+            .prepare(`SELECT session_id, event_subtype, duration_ms, metadata_json, created_at
+         FROM interaction_logs
+         WHERE session_id = ? AND event_type = 'ee_injection'
+         ORDER BY created_at DESC
          LIMIT ?`)
             .all(runId, limit);
     }
@@ -74,7 +74,7 @@ export function logInteraction(sessionId, eventType, metadata) {
     try {
         const db = getDatabase();
         const metadataJson = metadata?.data ? JSON.stringify(metadata.data) : null;
-        db.prepare(`INSERT INTO interaction_logs (session_id, event_type, event_subtype, model, duration_ms, input_tokens, output_tokens, metadata_json, created_at)
+        db.prepare(`INSERT INTO interaction_logs (session_id, event_type, event_subtype, model, duration_ms, input_tokens, output_tokens, metadata_json, created_at)
        VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?)`).run(sessionId, eventType, metadata?.eventSubtype ?? null, metadata?.model ?? null, metadata?.durationMs ?? null, metadata?.inputTokens ?? null, metadata?.outputTokens ?? null, metadataJson, new Date().toISOString());
         maybePruneOld();
     }