specrails-desktop 2.30.2 → 2.31.0

package/docs/tracking-cost.md

@@ -1,6 +1,6 @@
 # Tracking cost
 
-specrails-desktop records every AI CLI invocation it spawns — Claude, Codex, and Gemini — and surfaces the totals on one Analytics page per project. This guide walks through what's tracked, what's not, and how to read the dashboard.
+specrails-desktop records AI CLI invocations across Claude, Codex, Gemini, and Kimi and surfaces the available metrics on one Analytics page per project. This guide walks through what's tracked, what's unavailable, and how to read the dashboard.
 
 ## What gets tracked
 
@@ -15,7 +15,17 @@ Six surfaces, all per-project:
 | **`smash`** | SMASH runs that break an epic spec into sub-specs |
 | **`file-summary`** | Code-explorer AI summaries of individual files |
 
-Each invocation row carries: provider (Claude, Codex, or Gemini), model, status, started/finished timestamps, duration (wall clock + API-only), tokens (in / out / cache read / cache create), total USD cost, turn count, and — when applicable — the ticket and conversation IDs it touched.
+Each invocation row carries the provider, model, status, timestamps, duration,
+and relevant ticket/conversation IDs. Token, USD-cost, and turn fields are
+nullable because not every CLI reports them; Kimi leaves those values
+unavailable rather than writing zero.
+
+The surface list is app-wide, not a promise that every provider can spawn
+every row type. Kimi records its agentic chats/Explore/rails/loops, but Quick
+Spec, AI Edit, Contract Refine, SMASH/Re-SMASH, file summaries, construction
+story, AI auto-title, Project Builder blueprint/milestone generation, Loop
+Decider, and Agent Studio automation fail before spawn. Those rejected
+requests create no invocation row; AI auto-title uses a deterministic fallback.
 
 > Cost averages exclude `failed`/`aborted` rows, but those rows still count toward the total run count and the failure rate.
 
@@ -25,6 +35,9 @@ Whether a cost figure is exact depends on the provider's CLI, not on which provi
 
 - **Claude cost is provider-billed and authoritative** — the figure comes straight from the CLI's own usage report.
 - **Codex and Gemini do not report cost natively**, so the app **estimates** their cost from a local rate-card (`server/pricing.ts`) using the captured token counts. Estimated rows are flagged: they render with a `~` tilde in the raw table (hover for the tooltip — *"Estimated from local pricing table — this provider does not report cost natively"*) and feed an "includes ~$X estimated" footnote in the Hero.
+- **Kimi reports neither token counts nor a native USD-cost envelope** in its
+  stream. Specrails does not invent an estimate: cost and token cells stay
+  unavailable.
 
 On multi-provider projects, a **Provider breakdown** card splits spend across the project's installed engines so you can see authoritative vs estimated at a glance.
 
@@ -32,10 +45,12 @@ On multi-provider projects, a **Provider breakdown** card splits spend across th
 
 ## What's NOT tracked (intentionally)
 
-- **Sidebar chat** — the general-purpose chat panel in the right sidebar. It spawns an AI process but isn't pipeline work, so the app excludes it from analytics by design.
+- **Sidebar chat** — the general-purpose chat panel in the right sidebar. It spawns an AI process, including `kimi -p` when Kimi is selected, but isn't pipeline work, so the app excludes it from analytics by design.
 - **Setup wizard** — the install/enrich flow when you add a project. It *does* spawn an AI CLI (a genuine model invocation), but it's an interactive one-time wizard rather than a repeatable pipeline job, so it's deliberately left uninstrumented.
 
-If you want the absolute total of what an engine has cost you, your provider's own console is the source of truth — the Anthropic console for Claude, the OpenAI dashboard for Codex, and Google AI Studio / Cloud billing for Gemini.
+If you want the absolute total of what an engine has cost you, your provider's
+own console is the source of truth. This is especially important for Kimi,
+whose CLI stream provides no billable usage envelope to Desktop.
 
 ## The Analytics page
 
@@ -45,7 +60,7 @@ Open **Analytics** from the project right sidebar.
 
 - **Period** — `7d`, `30d`, `90d`, or `All`. (There's no 24-hour or custom-range option here — the custom calendar range lives on Desktop Analytics, below.)
 - **Surface** — chips for `All` plus each of the six surfaces (Jobs, Explore, Quick, Refine, SMASH, File summaries). All are included by default; toggle chips to narrow the view.
-- **Engine** — one chip per installed provider (Claude, Codex, Gemini) plus `All`. These appear **only on multi-provider projects** (more than one engine installed) and let you filter spend by engine.
+- **Engine** — one chip per installed provider plus `All`. These appear **only on multi-provider projects** and let you filter activity by engine.
 
 The period and surface filters are URL-synced, so you can share or bookmark a view.
 
@@ -62,7 +77,10 @@ When the project has zero invocations in the period (e.g. you just started), the
 
 ### Provider breakdown
 
-On multi-provider projects, a dedicated card splits total spend across the project's installed engines (Claude, Codex, Gemini) — it's data-driven, so it renders for any combination of more than one provider. Each engine's row marks an all-estimated total with a `~` tilde. It's hidden on single-provider projects.
+On multi-provider projects, a dedicated card splits available spend across the
+project's installed engines. It is data-driven, so Kimi appears even though its
+cost is unavailable. Estimated Codex/Gemini totals carry a `~` tilde. The card
+is hidden on single-provider projects.
 
 ### Daily stacked timeline
 
@@ -159,3 +177,4 @@ For the cross-project view, open **Analytics** from the Arc sidebar on the left.
 - [Creating specs](creating-specs.md) — every spec you create adds rows to your analytics.
 - [Using Codex](codex.md) — how cost works when you run a project on Codex (estimated, ~ tilde).
 - [Using Gemini](gemini.md) — how cost works when you run a project on Gemini (also estimated, ~ tilde).
+- [Using Kimi](kimi.md) — why Kimi duration/outcome are recorded while tokens and cost remain unavailable.

package/package.json

@@ -1,13 +1,13 @@
 {
   "name": "specrails-desktop",
-  "version": "2.30.2",
+  "version": "2.31.0",
   "license": "MIT",
   "repository": {
     "type": "git",
     "url": "git+https://github.com/fjpulidop/specrails-desktop.git"
   },
   "engines": {
-    "node": ">=20.0.0"
+    "node": ">=20.19.0"
   },
   "bin": {
     "specrails-desktop": "cli/dist/specrails-desktop.js"
@@ -51,6 +51,7 @@
     "cross-spawn": "^7.0.6",
     "diff": "^9.0.0",
     "express": "^5.2.1",
+    "js-yaml": "4.1.1",
     "multer": "^2.2.0",
     "node-pty": "^1.0.0",
     "pdf-parse": "^2.4.5",
@@ -69,6 +70,7 @@
     "@types/cross-spawn": "^6.0.6",
     "@types/diff": "^7.0.2",
     "@types/express": "^5.0.6",
+    "@types/js-yaml": "^4.0.9",
     "@types/multer": "^2.1.0",
     "@types/node": "^25.5.0",
     "@types/pdf-parse": "^1.1.5",

package/server/dist/agent-chat-manager.js

@@ -8,8 +8,10 @@ exports.sanitizeAgentTitle = sanitizeAgentTitle;
 const crypto_1 = require("crypto");
 const readline_1 = require("readline");
 const os_1 = require("os");
+const node_path_1 = __importDefault(require("node:path"));
 const tree_kill_1 = __importDefault(require("tree-kill"));
 const providers_1 = require("./providers");
+const runtime_1 = require("./providers/runtime");
 const spawn_lifecycle_1 = require("./spawn-lifecycle");
 const cli_prompt_1 = require("./util/cli-prompt");
 const result_event_1 = require("./result-event");
@@ -23,6 +25,18 @@ const attachment_manager_1 = require("./attachment-manager");
 const agent_context_resolver_1 = require("./agent-context-resolver");
 const agent_store_1 = require("./agent-store");
 const explore_draft_title_1 = require("./explore-draft-title");
+/**
+ * Providers report an expired or foreign resume token as a normalized error.
+ * Those errors are safe to recover from by clearing the token and retrying the
+ * same turn once; every other provider error remains terminal.
+ */
+function isStaleResumeError(message) {
+    if (!message)
+        return false;
+    return (/no rollout found for thread id/i.test(message) ||
+        /unknown (?:session|thread)(?: id)?\b/i.test(message) ||
+        /(?:session|thread)(?: id)?\b[^\n]{0,160}\b(?:not found|does not exist)\b/i.test(message));
+}
 class AgentChatManager {
     _broadcast;
     _db;
@@ -218,17 +232,20 @@ class AgentChatManager {
         // Resolve a model that is VALID for this provider. A stale model from another
         // provider (e.g. claude's "sonnet" after switching to codex) is rejected by
         // the provider — fall back to the provider's default instead of failing.
-        const catalog = new Set(adapter.modelCatalog().map((m) => m.value));
         const requested = options.model || conversation.model;
-        const model = requested && catalog.has(requested) ? requested : adapter.defaultModel();
-        // Reasoning effort: stored per conversation; null = the app default
-        // ("medium"). Passed only to providers with a per-spawn knob, and only when
-        // the value is in the provider's catalog (a provider switch clears it, but
-        // belt-and-braces against stale rows).
-        const efforts = (adapter.capabilities.reasoningEfforts ?? []);
+        const model = requested && (0, providers_1.isModelAvailableForAdapter)(adapter, requested)
+            ? requested
+            : adapter.defaultModel();
+        // Reasoning effort is model-scoped. A stale value from a previous model is
+        // discarded and the default is selected from the effective model's exact
+        // catalog (K3 follows its native high default; other Kimi models expose no
+        // per-invocation knob).
+        const efforts = (0, providers_1.reasoningEffortsForModel)(adapter, model);
         const storedEffort = conversation.reasoning_effort;
         const reasoningEffort = efforts.length
-            ? (storedEffort && efforts.includes(storedEffort) ? storedEffort : 'medium')
+            ? (storedEffort && efforts.includes(storedEffort)
+                ? storedEffort
+                : (0, providers_1.defaultReasoningEffortForModel)(adapter, model))
             : undefined;
         // Resolve attachments (conversation-keyed) into extracted text blocks +
         // absolute image paths. Extraction failure degrades to a text-only turn.
@@ -286,6 +303,9 @@ class AgentChatManager {
         const systemPrompt = hasAttachments
             ? `${agent_operator_prompt_1.OPERATOR_SYSTEM_PROMPT}\n\n${attachment_manager_1.USER_ATTACHMENT_SYSTEM_NOTE}`
             : agent_operator_prompt_1.OPERATOR_SYSTEM_PROMPT;
+        const effectivePrompt = adapter.capabilities.systemPromptArg
+            ? prompt
+            : `${systemPrompt}\n\n---\n\n${prompt}`;
         // Only pass native image paths to providers that can vision-load them (codex
         // `--image`); claude/gemini already have the `@path` ref folded into prompt.
         const spawnImagePaths = adapter.capabilities.supportsImageInput ? imagePaths : undefined;
@@ -300,7 +320,10 @@ class AgentChatManager {
             // / `.gemini/settings.json` and launch with the other's capability.
             // Claude and Codex keep their historical global cwd and explicit MCP
             // registration mechanisms unchanged.
-            const cwd = adapter.id === 'gemini'
+            const projectMcpPath = adapter.projectMcpPath?.('.');
+            const needsConversationMcpCwd = !!projectMcpPath &&
+                node_path_1.default.dirname(projectMcpPath) !== '.';
+            const cwd = needsConversationMcpCwd
                 ? (0, agent_cwd_manager_1.ensureAgentConversationCwd)(conversationId)
                 : (0, agent_cwd_manager_1.ensureAgentCwd)();
             let mcpArgs = [];
@@ -329,21 +352,22 @@ class AgentChatManager {
                 let capturedError = null;
                 console.log(`[agent-chat] turn start conv=${conversationId} provider=${adapter.id} binary=${adapter.binary} ` +
                     `action=${action} model=${model} cwd=${cwd} mcp=${mcpArgs.length ? 'mcp-config' : Object.keys(mcpEnv).join(',') || 'none'}`);
+                const buildOpts = {
+                    prompt: effectivePrompt,
+                    systemPrompt: adapter.capabilities.systemPromptArg ? systemPrompt : undefined,
+                    model,
+                    sessionId: useResume ? conversation.session_id ?? undefined : undefined,
+                    extraArgs: mcpArgs,
+                    imagePaths: spawnImagePaths,
+                    reasoning_effort: reasoningEffort,
+                };
                 const invocation = this._awaitWhileLive((0, spawn_lifecycle_1.runAiCliInvocation)({
                     adapter,
                     action,
                     cwd,
-                    env: { ...process.env, ...mcpEnv },
+                    env: (0, runtime_1.buildProviderEnv)(adapter, buildOpts, { ...process.env, ...mcpEnv }),
                     spawn: this._spawnOwned.bind(this),
-                    buildOpts: {
-                        prompt,
-                        systemPrompt: adapter.capabilities.systemPromptArg ? systemPrompt : undefined,
-                        model,
-                        sessionId: useResume ? conversation.session_id ?? undefined : undefined,
-                        extraArgs: mcpArgs,
-                        imagePaths: spawnImagePaths,
-                        reasoning_effort: reasoningEffort,
-                    },
+                    buildOpts,
                     onSpawn: (child) => {
                         if (this._disposed) {
                             this._terminate(child);
@@ -464,7 +488,10 @@ class AgentChatManager {
             // Auto-heal a stale/foreign session: a resume that produced no text (e.g. a
             // provider switch leaving another provider's session id, or codex
             // "no rollout found for thread id") retries once as a fresh turn.
-            if (canResume && !r.spawnFailed && !r.text) {
+            if (canResume &&
+                !r.spawnFailed &&
+                !r.text &&
+                (!r.error || isStaleResumeError(r.error))) {
                 console.log(`[agent-chat] resume produced no text — retrying fresh conv=${conversationId}`);
                 (0, agent_store_1.updateAgentConversation)(this._db, conversationId, { session_id: null });
                 r = await invoke(false);
@@ -485,6 +512,17 @@ class AgentChatManager {
                 record(r, 'failed');
                 return;
             }
+            // A normalized provider error is authoritative even when the CLI exits
+            // zero or streamed partial text before failing.
+            if (r.error || r.code !== 0) {
+                persistSession(null);
+                const reason = r.error ||
+                    (r.stderrTail ? r.stderrTail.split('\n').filter(Boolean).slice(-3).join(' ').slice(0, 300) : '') ||
+                    `${adapter.binary} exited with code ${r.code ?? 'unknown'}`;
+                this._emitError(conversationId, reason);
+                record(r, 'failed');
+                return;
+            }
             if (r.text) {
                 (0, agent_store_1.addAgentMessage)(this._db, { conversationId, role: 'assistant', content: r.text });
                 persistSession(r.sessionId);
@@ -499,9 +537,8 @@ class AgentChatManager {
             // Failure: reset the session so the NEXT turn starts fresh, and surface the
             // real reason instead of silently doing nothing.
             persistSession(null);
-            const reason = r.error ||
-                (r.stderrTail ? r.stderrTail.split('\n').filter(Boolean).slice(-3).join(' ').slice(0, 300) : '') ||
-                (r.code !== 0 ? `${adapter.binary} exited with code ${r.code}` : 'The agent returned no output.');
+            const reason = (r.stderrTail ? r.stderrTail.split('\n').filter(Boolean).slice(-3).join(' ').slice(0, 300) : '') ||
+                'The agent returned no output.';
             this._emitError(conversationId, reason);
             record(r, 'failed');
         }
@@ -523,8 +560,10 @@ class AgentChatManager {
         if (this._disposed)
             return;
         try {
+            const finishedAt = new Date().toISOString();
             const { result, estimated } = (0, result_event_1.finaliseInvocationResult)(adapter, outcome.events, {
                 fallbackModel: model,
+                durationMs: Math.max(0, Date.parse(finishedAt) - Date.parse(outcome.startedAt)),
             });
             (0, desktop_db_1.recordAgentInvocation)(this._db, {
                 id: (0, crypto_1.randomUUID)(),
@@ -533,7 +572,7 @@ class AgentChatManager {
                 provider: adapter.id,
                 status,
                 started_at: outcome.startedAt,
-                finished_at: new Date().toISOString(),
+                finished_at: finishedAt,
                 total_cost_usd_estimated: estimated,
                 ...result,
             });
@@ -614,13 +653,17 @@ class AgentChatManager {
         try {
             const conversationId = conversation.id;
             const adapter = (0, providers_1.getAdapter)(conversation.provider);
+            const toolPolicy = (0, runtime_1.pureOutputToolPolicy)(adapter);
+            if (!toolPolicy)
+                return;
             const model = adapter.defaultModel();
             const prompt = 'Generate a concise 3-6 word title for this conversation. ' +
                 'Output ONLY the title text — no quotes, no punctuation, no markdown, no preamble.\n\n' +
                 `User: ${firstUserMsg.slice(0, 240)}\nAssistant: ${firstResponse.slice(0, 320)}`;
-            const args = adapter.buildArgs('auto-title', { prompt, model });
+            const titleOptions = { prompt, model, toolPolicy };
+            const args = adapter.buildArgs('auto-title', titleOptions);
             const child = this._spawnOwned(adapter.binary, args, {
-                env: process.env,
+                env: (0, runtime_1.buildProviderEnv)(adapter, titleOptions),
                 stdio: ['ignore', 'pipe', 'pipe'],
                 cwd: (0, os_1.tmpdir)(),
             });
@@ -633,12 +676,11 @@ class AgentChatManager {
             reader.on('line', (line) => {
                 if (this._disposed)
                     return;
-                const ev = adapter.parseStreamLine(line);
-                if (!ev)
-                    return;
-                events.push(ev);
-                if (ev.kind === 'text-delta')
-                    raw += ev.text;
+                for (const ev of (0, runtime_1.parseStreamEvents)(adapter, line)) {
+                    events.push(ev);
+                    if (ev.kind === 'text-delta')
+                        raw += ev.text;
+                }
             });
             child.on('error', () => {
                 this._auxProcesses.delete(child);

package/server/dist/agent-chat-router.js

@@ -120,24 +120,30 @@ function createAgentChatRouter(deps) {
     router.get('/models', (req, res) => {
         const provider = validProvider(req.query.provider) ?? 'claude';
         const adapter = (0, providers_1.getAdapter)(provider);
+        const requestedModel = (0, providers_1.isModelAvailableForAdapter)(adapter, req.query.model)
+            ? req.query.model
+            : null;
+        const model = requestedModel ?? adapter.defaultModel();
         res.json({
             provider,
+            model,
             models: adapter.modelCatalog(),
+            customModelAliases: adapter.capabilities.customModelAliases === true,
             // The composer gates the image affordance on this (design D22: capability,
             // never provider id — gemini stays false until live-verified).
             supportsImageInput: adapter.capabilities.supportsImageInput === true,
             // Per-provider reasoning-effort tiers (ascending). Empty ⇒ no selector
             // (gemini has no per-spawn knob).
-            efforts: adapter.capabilities.reasoningEfforts ?? [],
+            efforts: (0, providers_1.reasoningEffortsForModel)(adapter, model),
         });
     });
-    /** Validate a requested reasoning effort against a provider's catalog. */
-    const validEffort = (provider, value) => {
+    /** Validate a requested effort against the effective model's exact catalog. */
+    const validEffort = (provider, model, value) => {
         if (value === null)
             return null;
         if (typeof value !== 'string')
             return undefined;
-        const efforts = (0, providers_1.getAdapter)(provider).capabilities.reasoningEfforts ?? [];
+        const efforts = (0, providers_1.reasoningEffortsForModel)((0, providers_1.getAdapter)(provider), model);
         return efforts.includes(value) ? value : undefined;
     };
     router.get('/conversations', (_req, res) => {
@@ -154,13 +160,20 @@ function createAgentChatRouter(deps) {
             }
             provider = v;
         }
+        const adapter = (0, providers_1.getAdapter)(provider);
+        const requestedModel = body.model == null ? null : body.model;
+        if (requestedModel !== null && !(0, providers_1.isModelAvailableForAdapter)(adapter, requestedModel)) {
+            res.status(400).json({ error: 'Model not available for this provider' });
+            return;
+        }
+        const effectiveModel = requestedModel ?? adapter.defaultModel();
         const conversation = (0, agent_store_1.createAgentConversation)(desktopDb, {
             provider,
-            model: typeof body.model === 'string' ? body.model : null,
+            model: requestedModel,
             pinnedProjectId: typeof body.pinnedProjectId === 'string' ? body.pinnedProjectId : null,
             tierLevel: body.tierLevel !== undefined ? (0, agent_tier_1.normalizeLevel)(body.tierLevel) : 0,
-            // Off-catalog values fall back to null (= app default "medium").
-            reasoningEffort: validEffort(provider, body.reasoningEffort) ?? null,
+            // Off-catalog values fall back to null (= no provider-specific override).
+            reasoningEffort: validEffort(provider, effectiveModel, body.reasoningEffort) ?? null,
         });
         res.status(201).json({ conversation });
     });
@@ -206,31 +219,41 @@ function createAgentChatRouter(deps) {
             }
         }
         // Explicit model pick — validated against the (new or current) provider's catalog.
-        if (body.model !== undefined && patch.model === undefined) {
+        if (body.model !== undefined) {
+            const effectiveProvider = patch.provider ?? conversation.provider;
             if (body.model === null) {
                 patch.model = null;
             }
             else if (typeof body.model === 'string') {
-                const effectiveProvider = patch.provider ?? conversation.provider;
-                const valid = new Set((0, providers_1.getAdapter)(effectiveProvider).modelCatalog().map((m) => m.value));
-                if (!valid.has(body.model)) {
+                if (!(0, providers_1.isModelAvailableForAdapter)((0, providers_1.getAdapter)(effectiveProvider), body.model)) {
                     res.status(400).json({ error: 'Model not available for this provider' });
                     return;
                 }
                 patch.model = body.model;
             }
         }
-        // Explicit effort pick — validated against the (new or current) provider's
-        // effort catalog; null resets to the app default ("medium").
-        if (body.reasoningEffort !== undefined && patch.reasoning_effort === undefined) {
-            const effectiveProvider = patch.provider ?? conversation.provider;
-            const v = validEffort(effectiveProvider, body.reasoningEffort);
+        const effectiveProvider = patch.provider ?? conversation.provider;
+        const adapter = (0, providers_1.getAdapter)(effectiveProvider);
+        const effectiveModel = typeof patch.model === 'string'
+            ? patch.model
+            : patch.model === null
+                ? adapter.defaultModel()
+                : conversation.model ?? adapter.defaultModel();
+        // Explicit effort pick — validated against the effective model. A model
+        // change without an effort pick clears a now-incompatible persisted value.
+        if (body.reasoningEffort !== undefined) {
+            const v = validEffort(effectiveProvider, effectiveModel, body.reasoningEffort);
             if (v === undefined) {
-                res.status(400).json({ error: 'Effort not available for this provider' });
+                res.status(400).json({ error: 'Effort not available for this provider and model' });
                 return;
             }
             patch.reasoning_effort = v;
         }
+        else if (body.model !== undefined
+            && conversation.reasoning_effort
+            && validEffort(effectiveProvider, effectiveModel, conversation.reasoning_effort) === undefined) {
+            patch.reasoning_effort = null;
+        }
         res.json({ conversation: (0, agent_store_1.updateAgentConversation)(desktopDb, conversation.id, patch) });
     });
     router.delete('/conversations/:id', (req, res) => {

package/server/dist/agent-cwd-manager.js

@@ -16,10 +16,10 @@ const agent_operator_prompt_1 = require("./agent-operator-prompt");
 //
 // The agent chat spawns the AI CLI from an app-owned cwd (NOT a project path), so
 // no project's CLAUDE.md is auto-loaded and the agent starts with a focused
-// "operator" stance. Claude/Codex retain the original app-global cwd. Gemini
-// gets a per-conversation child cwd because it discovers project MCP settings
-// from `<cwd>/.gemini/settings.json`; sharing that file would let concurrent
-// conversations overwrite each other's capability-file reference.
+// "operator" stance. Claude/Codex retain the original app-global cwd. Providers
+// whose MCP config is cwd-discovered under a nested project path (Gemini/Kimi)
+// get a per-conversation child cwd so concurrent turns cannot overwrite each
+// other's capability-file reference.
 function homeDir() {
     return process.env.SPECRAILS_REGISTRY_HOME || os_1.default.homedir();
 }
@@ -65,6 +65,14 @@ function ensureAgentCwd() {
         fs_1.default.rmdirSync(path_1.default.join(dir, '.gemini'));
     }
     catch { /* non-empty / absent */ }
+    try {
+        fs_1.default.unlinkSync(path_1.default.join(dir, '.kimi-code', 'mcp.json'));
+    }
+    catch { /* absent */ }
+    try {
+        fs_1.default.rmdirSync(path_1.default.join(dir, '.kimi-code'));
+    }
+    catch { /* non-empty / absent */ }
     return dir;
 }
 /** Path for provider state that must not be shared across agent conversations. */
@@ -75,9 +83,8 @@ function agentConversationCwdPath(conversationId) {
     return path_1.default.join(agentCwdPath(), 'conversations', conversationId);
 }
 /**
- * Materialise an isolated agent cwd for one conversation. Gemini uses this so
- * its cwd-discovered `.mcp.json` and `.gemini/settings.json` can never be
- * overwritten by a concurrent turn belonging to another conversation.
+ * Materialise an isolated agent cwd for one conversation. Cwd-discovered
+ * provider MCP files can never be overwritten by another conversation.
  */
 function ensureAgentConversationCwd(conversationId) {
     return ensureCwd(agentConversationCwdPath(conversationId));