codeep 3.3.0 → 3.3.2

package/dist/config/index.d.ts

@@ -183,6 +183,19 @@ export declare function initializeAsProject(path: string): boolean;
  * Check if folder was manually initialized as project
  */
 export declare function isManuallyInitializedProject(path: string): boolean;
+/**
+ * Create config with fallback logic
+ * 1. Try standard Conf location (~/.config/codeep-nodejs on Linux, etc.)
+ * 2. If not writable, use .codeep in current working directory
+ */
+/**
+ * What a fresh install starts on. Exported so a test can hold the pair to the
+ * catalogue: the model had stayed `glm-5.2` after Z.AI's default moved to
+ * `glm-5.3`, so new users started a flagship behind what the website promised,
+ * and nothing failed because 5.2 still works.
+ */
+export declare const DEFAULT_PROVIDER = "z.ai";
+export declare const DEFAULT_MODEL = "glm-5.3";
 export declare const config: Conf<ConfigSchema>;
 export declare const LANGUAGES: Record<string, string>;
 export declare const PROTOCOLS: Record<string, string>;

package/dist/config/index.js

@@ -138,12 +138,20 @@ function isWritable(dir) {
  * 1. Try standard Conf location (~/.config/codeep-nodejs on Linux, etc.)
  * 2. If not writable, use .codeep in current working directory
  */
+/**
+ * What a fresh install starts on. Exported so a test can hold the pair to the
+ * catalogue: the model had stayed `glm-5.2` after Z.AI's default moved to
+ * `glm-5.3`, so new users started a flagship behind what the website promised,
+ * and nothing failed because 5.2 still works.
+ */
+export const DEFAULT_PROVIDER = 'z.ai';
+export const DEFAULT_MODEL = 'glm-5.3';
 function createConfig() {
     const defaults = {
         apiKey: '',
         migrationVersion: 0,
-        provider: 'z.ai',
-        model: 'glm-5.2',
+        provider: DEFAULT_PROVIDER,
+        model: DEFAULT_MODEL,
         agentMode: 'on',
         ollamaUrl: 'http://localhost:11434',
         ollamaNativeApi: false,

package/dist/renderer/commands/helpers.d.ts

@@ -85,6 +85,10 @@ export interface StatsCache {
     cacheReadTokens: number;
     cacheCreationTokens: number;
     estimatedSavingsUsd: number;
+    /** Rates that applied, from getCacheStats. Absent → no rate is quoted rather
+     *  than assuming Anthropic's 0.1×, which was wrong for DeepSeek, Kimi, Qwen
+     *  and Fable 5.1. */
+    cacheReadRates?: number[];
 }
 export interface PricingRow {
     model: string;

package/dist/renderer/commands/helpers.js

@@ -8,6 +8,7 @@
  * coverage.
  */
 import { isFlatFeeProvider } from '../../config/providers.js';
+import { formatCacheReadRates } from '../../utils/tokenTracker.js';
 /** Snippet window: chars of context before / after the match. */
 export const SEARCH_SNIPPET_BEFORE = 30;
 export const SEARCH_SNIPPET_AFTER = 50;
@@ -166,7 +167,7 @@ export function formatStatsReport(args) {
         }
         if (cache.cacheReadTokens > 0 || cache.cacheCreationTokens > 0) {
             lines.push('', '### Prompt caching');
-            lines.push(`Cache reads: ${fmt(cache.cacheReadTokens)} tokens (billed at 0.1× input rate)`);
+            lines.push(`Cache reads: ${fmt(cache.cacheReadTokens)} tokens${formatCacheReadRates(cache.cacheReadRates ?? [])}`);
             if (cache.cacheCreationTokens > 0) {
                 lines.push(`Cache writes: ${fmt(cache.cacheCreationTokens)} tokens (billed at 1.25× input rate)`);
             }

package/dist/renderer/commands/registry.d.ts

@@ -1,11 +1,24 @@
 /**
- * Single source of truth for Codeep slash commands.
+ * Single source of truth for Codeep's TUI slash commands.
  *
- * Every other place that needs command metadata — the `/` autocomplete in
- * `App.ts`, the `/help` screen in `components/Help.ts`, the dispatcher in
- * `commands.ts`, and the ACP command handler in `acp/commands.ts` — derives
- * from this registry. Adding a command means adding one entry here; the
- * autocomplete list, help screen, and command index all pick it up.
+ * The `/` autocomplete in `App.ts` and alias resolution in the dispatcher
+ * (`commands.ts`) derive from this registry. Adding a command means:
+ *
+ * 1. an entry in `COMMANDS` below;
+ * 2. a handler — a `case` in `renderer/commands.ts`, a `case` in `App.ts`, or
+ *    a built-in skill of the same name;
+ * 3. a row in `HELP_LAYOUT` at the bottom of this file;
+ * 4. `npm run export:commands`, which regenerates the website's command
+ *    reference (`Codeep-web/src/data/commands.json`) from this file.
+ *
+ * `registry.test.ts` fails if 2 or 3 is missed: a command offered in the
+ * autocomplete with no handler answers "Unknown command" (this is how
+ * `/account` shipped broken), and one missing from `HELP_LAYOUT` is invisible
+ * in `/help` and on the website.
+ *
+ * The ACP server (VS Code, Zed) is NOT derived from here: it keeps its own
+ * `AVAILABLE_COMMANDS` list in `acp/server.ts` and resolves its own aliases,
+ * so a command added here is not automatically offered to editors.
  *
  * ## What lives here vs. elsewhere
  *
@@ -76,7 +89,11 @@ export declare const ALL_ALIASES: ReadonlySet<string>;
 export interface HelpItemSpec {
     /** Visible key in `/help`, including the leading `/`. */
     key: string;
+    /** One line, sized for the terminal. */
     description: string;
+    /** Longer explanation used on the website's command reference instead of
+     *  `description`, where a terminal row is too short to say what matters. */
+    web?: string;
 }
 export interface HelpCategorySpec {
     title: string;

package/dist/renderer/commands/registry.js

@@ -1,11 +1,24 @@
 /**
- * Single source of truth for Codeep slash commands.
+ * Single source of truth for Codeep's TUI slash commands.
  *
- * Every other place that needs command metadata — the `/` autocomplete in
- * `App.ts`, the `/help` screen in `components/Help.ts`, the dispatcher in
- * `commands.ts`, and the ACP command handler in `acp/commands.ts` — derives
- * from this registry. Adding a command means adding one entry here; the
- * autocomplete list, help screen, and command index all pick it up.
+ * The `/` autocomplete in `App.ts` and alias resolution in the dispatcher
+ * (`commands.ts`) derive from this registry. Adding a command means:
+ *
+ * 1. an entry in `COMMANDS` below;
+ * 2. a handler — a `case` in `renderer/commands.ts`, a `case` in `App.ts`, or
+ *    a built-in skill of the same name;
+ * 3. a row in `HELP_LAYOUT` at the bottom of this file;
+ * 4. `npm run export:commands`, which regenerates the website's command
+ *    reference (`Codeep-web/src/data/commands.json`) from this file.
+ *
+ * `registry.test.ts` fails if 2 or 3 is missed: a command offered in the
+ * autocomplete with no handler answers "Unknown command" (this is how
+ * `/account` shipped broken), and one missing from `HELP_LAYOUT` is invisible
+ * in `/help` and on the website.
+ *
+ * The ACP server (VS Code, Zed) is NOT derived from here: it keeps its own
+ * `AVAILABLE_COMMANDS` list in `acp/server.ts` and resolves its own aliases,
+ * so a command added here is not automatically offered to editors.
  *
  * ## What lives here vs. elsewhere
  *
@@ -67,13 +80,12 @@ export const COMMANDS = [
     { name: 'settings', description: 'Open settings', category: 'general' },
     { name: 'version', description: 'Show version', category: 'general' },
     { name: 'update', description: 'Check for updates', category: 'general' },
-    {
-        name: 'cost',
-        aliases: ['stats'],
-        description: 'Token usage & cost this session',
-        category: 'general',
-        aliasListed: true,
-    },
+    { name: 'cost', description: 'Token usage & cost this session', category: 'general' },
+    // Its own command, not an alias of /cost. From 2026-06-24 it was
+    // `aliases: ['stats']` on /cost, and the dispatcher resolves aliases before
+    // the switch — so /stats printed the /cost report and its own handler (per-
+    // model breakdown, cache summary, pricing table) never ran.
+    { name: 'stats', description: 'Detailed session view — per-model breakdown, cache, and pricing table', category: 'general' },
     { name: 'clear', description: 'Clear chat', category: 'general', hidden: true },
     { name: 'exit', description: 'Quit application', category: 'general', hidden: true },
     // ── sessions ───────────────────────────────────────────────────────────────
@@ -220,7 +232,7 @@ export const COMMANDS = [
     { name: 'commands', description: 'List custom slash commands in .codeep/commands/*.md', category: 'extensions' },
     { name: 'web-cache', aliases: ['webcache'], description: 'Show @web fetch cache stats (alias: /web-cache clear)', category: 'extensions' },
     // ── cloud & account ────────────────────────────────────────────────────────
-    { name: 'account', description: 'Link this machine to your codeep.dev account', category: 'cloud' },
+    { name: 'account', description: 'Show whether this machine is linked to codeep.dev, and how to link it', category: 'cloud' },
     { name: 'tasks', description: 'List/add/done/delete codeep.dev tasks', category: 'cloud', usage: ['add <title> [--bug|--feature]'] },
     { name: 'sync', description: 'Sync learning preferences and profiles to codeep.dev', category: 'cloud' },
     { name: 'telemetry', description: 'Show or toggle automatic cloud telemetry (on/off)', category: 'cloud' },
@@ -229,7 +241,11 @@ export const COMMANDS = [
     { name: 'context-save', description: 'Save conversation', category: 'cloud', hidden: true },
     { name: 'context-load', description: 'Load conversation', category: 'cloud', hidden: true },
     { name: 'context-clear', description: 'Clear saved context', category: 'cloud', hidden: true },
-    { name: 'save', description: 'Save current session', category: 'cloud', hidden: true },
+    // ACP-only. acp/commands.ts handles /save (an editor has no autosave); the TUI
+    // saves sessions on its own and has no handler, so this stays hidden — offering
+    // it in the terminal would only answer "Unknown command". Registered because
+    // registry.test.ts requires every ACP case label to be.
+    { name: 'save', description: 'Save current session (VS Code / Zed only)', category: 'cloud', hidden: true },
     // ── code generation ────────────────────────────────────────────────────────
     { name: 'build', description: 'Build the project', category: 'codegen' },
     { name: 'deploy', description: 'Build and deploy', category: 'codegen' },
@@ -333,7 +349,10 @@ export const HELP_LAYOUT = [
             { key: '/settings', description: 'Open settings' },
             { key: '/version', description: 'Show version' },
             { key: '/update', description: 'Check for updates' },
-            { key: '/stats (/cost)', description: 'Token usage & cost this session' },
+            { key: '/cost', description: 'Token usage & cost this session',
+                web: 'Token usage and estimated cost this session, per provider and model, with prompt-cache savings' },
+            { key: '/stats', description: 'Detailed view — per-model breakdown, cache, pricing table',
+                web: 'Detailed session view — per-model breakdown, plan-aware totals, prompt-cache reads and writes, and the per-1M pricing table' },
             { key: '/clear', description: 'Clear chat' },
             { key: '/exit', description: 'Quit application' },
         ],
@@ -345,11 +364,15 @@ export const HELP_LAYOUT = [
             { key: '/new', description: 'Start new session' },
             { key: '/rename <name>', description: 'Rename current session' },
             { key: '/search <term>', description: 'Search the current session' },
-            { key: '/recall <query>', description: 'Search across ALL saved sessions (cross-session)' },
-            { key: '/recall … --resume', description: 'Load the top-matching session directly' },
+            { key: '/recall <query>', description: 'Search across ALL saved sessions (cross-session)',
+                web: 'Search across all saved sessions, ranked by relevance and recency' },
+            { key: '/recall … --resume', description: 'Load the top-matching session directly',
+                web: 'Load the top-matching session directly, skipping the picker' },
             { key: '/recall … --summarize', description: 'LLM recap of what you did across matches' },
+            { key: '/cloud', description: 'List and resume sessions synced from other devices' },
             { key: '/export [md|json|txt]', description: 'Export chat' },
-            { key: '/compact [keepN]', description: 'AI-summarize older messages to free up context (keeps last N)' },
+            { key: '/compact [keepN]', description: 'AI-summarize older messages to free up context (keeps last N)',
+                web: 'AI-summarize older messages to free up context, keeping the last N (default 4)' },
         ],
     },
     {
@@ -357,7 +380,8 @@ export const HELP_LAYOUT = [
         items: [
             { key: '/checkpoint [name]', description: 'Snapshot conversation + provider/model + git HEAD' },
             { key: '/checkpoints', description: 'List saved checkpoints in this workspace' },
-            { key: '/rewind <id>', description: 'Restore conversation from a checkpoint' },
+            { key: '/rewind <id>', description: 'Restore conversation from a checkpoint',
+                web: 'Restore a checkpoint — conversation and provider/model; files roll back via the git command printed afterwards' },
             { key: '/checkpoint delete <id>', description: 'Delete a saved checkpoint' },
         ],
     },
@@ -387,6 +411,9 @@ export const HELP_LAYOUT = [
             { key: '/amend', description: 'Amend last commit' },
             { key: '/branch', description: 'Create/manage branches' },
             { key: '/stash', description: 'Stash changes' },
+            { key: '/unstash', description: 'Apply and drop the most recent stash' },
+            { key: '/pr', description: 'Create a pull request description' },
+            { key: '/changelog', description: 'Generate changelog from recent commits' },
             { key: '/init', description: 'Initialize project (.codeep/ folder)' },
             { key: '/scan', description: 'Scan project structure' },
             { key: '/memory <note>', description: 'Add note to project intelligence' },
@@ -414,19 +441,58 @@ export const HELP_LAYOUT = [
         title: 'Skills (Shortcuts)',
         items: [
             { key: '/test (/t)', description: 'Generate/run tests' },
+            { key: '/test-fix', description: 'Fix failing tests' },
+            { key: '/coverage', description: 'Run/analyze test coverage' },
+            { key: '/e2e', description: 'Generate end-to-end tests' },
+            { key: '/mock', description: 'Generate mock data for testing' },
             { key: '/docs (/d)', description: 'Open web docs for a command' },
+            { key: '/readme', description: 'Generate or update README' },
+            { key: '/api-docs', description: 'Generate API documentation' },
             { key: '/refactor (/r)', description: 'Improve code quality' },
             { key: '/fix (/f)', description: 'Debug and fix issues' },
             { key: '/explain (/e)', description: 'Explain code' },
             { key: '/optimize (/o)', description: 'Optimize performance' },
             { key: '/debug (/b)', description: 'Debug problems' },
             { key: '/security', description: 'Security audit (SQLi, XSS, secrets, auth)' },
-            { key: '/coverage', description: 'Run/analyze test coverage' },
+            { key: '/types', description: 'Add or improve TypeScript types' },
+            { key: '/cleanup', description: 'Clean up code (remove unused, format)' },
+            { key: '/modernize', description: 'Update code to use modern syntax' },
+            { key: '/migrate', description: 'Migrate code to newer version' },
+            { key: '/split', description: 'Split a large file into smaller modules' },
+            { key: '/log', description: 'Add logging to code' },
+            { key: '/translate', description: 'Translate code comments to English' },
+            { key: '/skills', description: 'List all skills' },
+            { key: '/skills <query>', description: 'Search skills by keyword' },
+            { key: '/skills bundles', description: 'List skill bundles (project + global)' },
+            { key: '/skills create-bundle <name>', description: 'Scaffold a bundle in .codeep/skills/' },
+            { key: '/skills show <name>', description: "Print a bundle's SKILL.md" },
+            { key: '/skills browse [query]', description: 'Search the marketplace at codeep.dev/skills' },
+            { key: '/skills install <owner>/<slug>', description: 'Install a marketplace bundle into the project' },
+            { key: '/skills publish <name> [--public]', description: 'Share a bundle to codeep.dev' },
+            { key: '/skills unpublish <owner>/<slug>', description: 'Remove a bundle you published' },
+        ],
+    },
+    {
+        title: 'Code Generation',
+        items: [
             { key: '/component <name>', description: 'Generate UI component' },
             { key: '/api <name>', description: 'Generate API endpoint' },
+            { key: '/hook', description: 'Generate a React hook' },
+            { key: '/service', description: 'Generate a service/utility module' },
+            { key: '/page', description: 'Generate a new page/route' },
+            { key: '/form', description: 'Generate a form with validation' },
+            { key: '/crud', description: 'Generate full CRUD for an entity' },
             { key: '/docker', description: 'Generate Dockerfile + compose' },
-            { key: '/skills', description: 'List all 50+ skills' },
-            { key: '/skills <query>', description: 'Search skills by keyword' },
+            { key: '/ci', description: 'Generate CI/CD configuration' },
+            { key: '/env', description: 'Setup environment configuration' },
+            { key: '/k8s', description: 'Generate Kubernetes manifests' },
+            { key: '/terraform', description: 'Generate Terraform configuration' },
+            { key: '/nginx', description: 'Generate Nginx configuration' },
+            { key: '/monitor', description: 'Add monitoring and observability' },
+            { key: '/build', description: 'Build the project' },
+            { key: '/deploy', description: 'Build and deploy' },
+            { key: '/release', description: 'Create a new release' },
+            { key: '/publish', description: 'Publish package to npm' },
         ],
     },
     {
@@ -441,16 +507,33 @@ export const HELP_LAYOUT = [
             { key: '/login', description: 'Login with API key' },
             { key: '/logout', description: 'Logout from provider' },
             { key: '/profile save <name>', description: 'Save current provider+model as profile' },
+            { key: '/profile load <name>', description: 'Load a saved profile' },
             { key: '/profile list', description: 'List saved profiles' },
-            { key: '/openrouter', description: 'OpenRouter routing prefs (prefer/ignore providers, fallbacks, privacy)' },
+            { key: '/profile delete <name>', description: 'Delete a saved profile' },
+            { key: '/openrouter', description: 'OpenRouter routing prefs (prefer/ignore providers, fallbacks, privacy)',
+                web: 'OpenRouter routing preferences: show, prefer or ignore upstream providers, fallbacks on/off, privacy strict/allow, clear' },
             { key: '/personality', description: 'List personalities and custom bots with model, tools, and scope' },
             { key: '/personality <name>', description: 'Activate a personality or custom bot. /personality off to clear.' },
-            { key: '/me', description: 'Your user profile (reply language, style, stack) — adapts the agent to you' },
+            { key: '/me', description: 'Your user profile (reply language, style, stack) — adapts the agent to you',
+                web: 'Your user profile — reply language, style, stack — injected into every run on every surface' },
             { key: '/me init [project]', description: 'Scaffold a profile template (global, or for this project). /me off to disable' },
             { key: '/me learn [on|off]', description: 'Learn durable prefs from this session now; on/off toggles auto-learn. /me forget clears it' },
             { key: '/me sync', description: 'Push your profile to the codeep.dev dashboard (and pull on a fresh machine)' },
-            { key: '/agents', description: 'List sub-agents the agent can delegate to (researcher / reviewer / tester / your own)' },
+            { key: '/agents', description: 'List sub-agents the agent can delegate to (researcher / reviewer / tester / your own)',
+                web: 'List the sub-agents the agent can delegate to — planner, researcher, reviewer, tester, or your own in .codeep/agents/' },
             { key: '/insights [--days N]', description: 'Activity summary — runs, files, tools, projects over the last N days (default 7)' },
+            { key: '/telegram', description: 'Set up answering confirmations on your phone' },
+            { key: '/telegram on|off', description: 'Turn Telegram approvals on or off' },
+            { key: '/audit', description: 'What agents did here — runs, tools, and what the boundary refused' },
+            { key: '/audit on|off', description: 'Turn the project audit record on or off' },
+        ],
+    },
+    {
+        title: 'Thinking',
+        items: [
+            { key: '/thinking (/effort)', description: 'Show the thinking tier and what this model supports' },
+            { key: '/thinking <auto|low|medium|high|max>', description: 'Set how hard the model reasons',
+                web: "Set the reasoning-effort tier. Auto uses each model's own default; other tiers are clamped to the levels the current provider and model actually distinguish, so an unsupported value is never sent. Models with no graded control ignore it." },
         ],
     },
     {
@@ -468,6 +551,25 @@ export const HELP_LAYOUT = [
             { key: '/mcp prompt <server> <name>', description: 'Materialize a prompt with arguments (key=value)' },
             { key: '/hooks', description: 'List installed lifecycle hooks (.codeep/hooks/<event>.sh)' },
             { key: '/commands', description: 'List custom slash commands (.codeep/commands/*.md)' },
+            { key: '/web-cache', description: 'Show @web fetch cache stats' },
+            { key: '/web-cache clear', description: 'Clear the @web fetch cache' },
+        ],
+    },
+    {
+        title: 'Cloud & Account',
+        items: [
+            { key: '/account', description: 'Is this machine linked to codeep.dev? How to link it',
+                web: 'Show whether this machine is linked to codeep.dev. Linking runs in your shell as `codeep account`, and the approval link it prints can be opened on any device where you are already signed in' },
+            { key: '/tasks', description: 'List codeep.dev tasks' },
+            { key: '/tasks add <title> [--bug|--feature] [--desc <text>]', description: 'Create a task on the dashboard' },
+            { key: '/tasks done <n>', description: 'Mark task #n as done' },
+            { key: '/tasks delete <n>', description: 'Delete task #n' },
+            { key: '/sync', description: 'Sync learning preferences and profiles to codeep.dev' },
+            { key: '/sync learning', description: 'Sync only learning preferences' },
+            { key: '/sync profiles', description: 'Sync only saved profiles' },
+            { key: '/telemetry [on|off]', description: 'Show or toggle automatic cloud telemetry' },
+            { key: '/keysync [on|off]', description: 'Show or toggle syncing API keys to codeep.dev',
+                web: 'Show or toggle syncing API keys to codeep.dev. Off by default — keys stay in your OS keychain unless you opt in' },
         ],
     },
     {

package/dist/renderer/commands.js

@@ -288,7 +288,6 @@ export async function handleCommand(command, args, ctx) {
             }
             break;
         }
-        case 'effort':
         case 'thinking': {
             const providerId = config.get('provider');
             const model = config.get('model');
@@ -302,7 +301,7 @@ export async function handleCommand(command, args, ctx) {
                     ctx.app.notify('Thinking effort: auto — each model uses its own default.');
                 }
                 else if (!supported) {
-                    ctx.app.notify(`Thinking effort set to "${sub}", but ${model} has no graded thinking control — it will be ignored until you switch to a model that does (e.g. Opus 5, GPT-5.x, Gemini 3, DeepSeek V4, Kimi K3).`);
+                    ctx.app.notify(`Thinking effort set to "${sub}", but ${model} has no graded thinking control — it will be ignored until you switch to a model that does (e.g. Opus 5, GPT-5.x or GPT-6, Gemini 3, DeepSeek V4.1 Flash, GLM-5.x, Kimi K3).`);
                 }
                 else {
                     // Tell the user what THIS model will actually run (the tier may
@@ -377,7 +376,6 @@ export async function handleCommand(command, args, ctx) {
             runAgentTask(args.join(' '), true, ctx, () => null, () => { });
             break;
         }
-        case 'd':
         case 'docs': {
             // Open per-command web docs in the system browser. Lets the inline
             // /help stay terse (single-line entries) while users who want the
@@ -1771,21 +1769,13 @@ Format: use headers per category, only include categories where you found issues
             break;
         }
         // Built-in skill shortcuts
-        case 'c':
         case 'commit':
-        case 't':
         case 'test':
-        case 'r':
         case 'refactor':
-        case 'f':
         case 'fix':
-        case 'e':
         case 'explain':
-        case 'o':
         case 'optimize':
-        case 'b':
         case 'debug':
-        case 'p':
         case 'push':
         case 'pull':
         case 'amend':
@@ -2025,6 +2015,23 @@ Describe what this skill does. The agent reads this body verbatim when it invoke
             });
             break;
         }
+        case 'account': {
+            // Linking cannot run here: `codeep account` prints to stdout and exits the
+            // process when it finishes, which would tear down the TUI. Before this
+            // case existed the autocomplete offered /account and it answered
+            // "Unknown command" — so this says where the flow runs, and what state
+            // the machine is in now.
+            const { getSyncToken, getGithubId } = await import('../config/index.js');
+            const linked = Boolean(getSyncToken());
+            const githubId = getGithubId();
+            ctx.app.addMessage({
+                role: 'system',
+                content: linked
+                    ? `## codeep.dev account\n\nThis machine is **linked**${githubId ? ` (GitHub id ${githubId})` : ''}. Usage and sessions sync to your dashboard.\n\nTo cut it off, use **Settings → Connected devices** on codeep.dev.`
+                    : `## codeep.dev account\n\nThis machine is **not linked**.\n\nQuit Codeep and run this in your shell:\n\n\`\`\`bash\ncodeep account\n\`\`\`\n\nIt prints a one-time approval link. You can open it on **any** device where you are already signed in to codeep.dev — it does not have to be this machine, so there is no need to sign in to GitHub here.`,
+            });
+            break;
+        }
         case 'tasks': {
             const { fetchTasks, markTaskDone, generateProjectId } = await import('../utils/codeepCloud.js');
             const { setTaskContext, clearTaskContext } = await import('../utils/taskContext.js');

package/dist/utils/tokenTracker.d.ts

@@ -93,6 +93,21 @@ export interface ProviderCostBreakdown {
     cacheReadTokens: number;
     estimatedCost: number;
 }
+/**
+ * What a cached prompt token costs, as a fraction of the model's input rate.
+ * Model first (a property of the model), then provider, then the default.
+ *
+ * One lookup for everything that needs it. Cost used this chain while savings
+ * hardcoded 0.1, so every provider priced differently from Anthropic had a
+ * cost and a "saved" figure that disagreed with each other.
+ */
+export declare function cacheReadRateFor(model: string, provider: string | undefined): number;
+/**
+ * The rate note for a report. One rate reads as "0.02×"; a session mixing
+ * providers reads as a range, because any single number there would be wrong
+ * for part of it.
+ */
+export declare function formatCacheReadRates(rates: readonly number[]): string;
 /**
  * Get cost breakdown grouped by provider/model.
  *
@@ -120,6 +135,9 @@ export interface CacheStats {
     /** True when EVERY cached token came from a flat-fee plan — there is no
      *  metered spend to have saved against. */
     isEntirelyFlatFeeCache: boolean;
+    /** The read rate of each metered record that read from cache, so a report
+     *  can state the rate that actually applied instead of assuming 0.1×. */
+    cacheReadRates: number[];
 }
 export declare function getCacheStats(): CacheStats;
 /**

package/dist/utils/tokenTracker.js

@@ -127,9 +127,10 @@ const MODEL_PRICING = {
     // rate, so it carries PEAK — an over-estimate by design. The previous rows
     // (0.435/0.87 and 0.14/0.28) were two to four and a half times below today's
     // peak and so under-reported, which is the one direction this table must not
-    // err in. Cache hits cost 2% of a miss, but DeepSeek reports them as
-    // `prompt_cache_hit_tokens`, which this tracker does not read — so every input
-    // token bills at the miss rate: an over-estimate again, not a gap.
+    // err in. Cache hits are read (DeepSeek reports them both nested as
+    // `prompt_tokens_details.cached_tokens` and top-level as
+    // `prompt_cache_hit_tokens`) and billed at DeepSeek's own hit rate — see
+    // CACHE_READ_RATE below.
     'deepseek-flash': { inputPer1M: 0.30, outputPer1M: 1.20 },
     // Retired V4 Flash is served by V4.1 Flash and billed at its price.
     'deepseek-v4-flash': { inputPer1M: 0.30, outputPer1M: 1.20 },
@@ -250,8 +251,12 @@ export function extractOpenAIUsage(data) {
         // later. Reading only the nested form zeroed every Kimi cache hit, so the
         // cached portion of a run billed at the full cache-miss rate — five times
         // what it costs — with nothing anywhere to say so.
+        // DeepSeek sends the same number twice — nested `cached_tokens` and
+        // top-level `prompt_cache_hit_tokens` — so the nested read already covers
+        // it; the top-level field is a last resort, never added to the other.
         const nested = data.usage.prompt_tokens_details?.cached_tokens;
-        const cached = (typeof nested === 'number' ? nested : data.usage.cached_tokens) || 0;
+        const topLevel = data.usage.cached_tokens ?? data.usage.prompt_cache_hit_tokens;
+        const cached = (typeof nested === 'number' ? nested : topLevel) || 0;
         return {
             promptTokens: data.usage.prompt_tokens || 0,
             completionTokens: data.usage.completion_tokens || 0,
@@ -301,8 +306,16 @@ export function extractAnthropicUsage(data) {
  */
 const MODEL_CACHE_READ_RATE = {
     'claude-fable-5-1': 0.025,
+    // V4 Pro's own ratio: $0.044 hit against $1.32 miss (peak; off-peak halves
+    // both, so the ratio holds). Historical — V4 Pro routes to V4.1 Flash from
+    // 2026-09-14 and configs holding it are migrated.
+    'deepseek-v4-pro': 0.044 / 1.32,
 };
 const CACHE_READ_RATE = {
+    // V4.1 Flash: $0.006 hit against $0.30 miss at peak, $0.003 against $0.15
+    // off-peak — 0.02 either way. Without an entry DeepSeek fell to the 0.1
+    // default and every cached token billed at five times its price.
+    'deepseek': 0.02,
     'kimi': 0.2,
     'kimi-api': 0.2,
     'qwen': 0.2,
@@ -313,6 +326,33 @@ const CACHE_READ_RATE = {
 };
 /** Anthropic's ratio, and the safest guess for a provider we have not priced. */
 const DEFAULT_CACHE_READ_RATE = 0.1;
+/**
+ * What a cached prompt token costs, as a fraction of the model's input rate.
+ * Model first (a property of the model), then provider, then the default.
+ *
+ * One lookup for everything that needs it. Cost used this chain while savings
+ * hardcoded 0.1, so every provider priced differently from Anthropic had a
+ * cost and a "saved" figure that disagreed with each other.
+ */
+export function cacheReadRateFor(model, provider) {
+    return MODEL_CACHE_READ_RATE[model]
+        ?? CACHE_READ_RATE[provider?.trim().toLowerCase() ?? '']
+        ?? DEFAULT_CACHE_READ_RATE;
+}
+/**
+ * The rate note for a report. One rate reads as "0.02×"; a session mixing
+ * providers reads as a range, because any single number there would be wrong
+ * for part of it.
+ */
+export function formatCacheReadRates(rates) {
+    const fmt = (r) => `${Number(r.toFixed(3))}×`;
+    const distinct = [...new Set(rates.map(r => Number(r.toFixed(4))))].sort((a, b) => a - b);
+    if (distinct.length === 0)
+        return '';
+    if (distinct.length === 1)
+        return ` (billed at ${fmt(distinct[0])} input rate)`;
+    return ` (billed at ${fmt(distinct[0])}–${fmt(distinct[distinct.length - 1])} input rate, by model)`;
+}
 /**
  * Get cost breakdown grouped by provider/model.
  *
@@ -346,9 +386,7 @@ export function getCostBreakdown(startIndex = 0) {
                 // prompt tokens bill at the standard 1.0× rate.
                 const cacheCreate = record.cacheCreationTokens ?? 0;
                 const cacheRead = record.cacheReadTokens ?? 0;
-                const cacheReadRate = MODEL_CACHE_READ_RATE[record.model]
-                    ?? CACHE_READ_RATE[record.provider?.trim().toLowerCase()]
-                    ?? DEFAULT_CACHE_READ_RATE;
+                const cacheReadRate = cacheReadRateFor(record.model, record.provider);
                 const uncachedPrompt = Math.max(0, record.promptTokens - cacheCreate - cacheRead);
                 existing.estimatedCost +=
                     (uncachedPrompt / 1_000_000) * pricing.inputPer1M
@@ -367,6 +405,7 @@ export function getCacheStats() {
     let savings = 0;
     let flatFeeCached = 0;
     let meteredCached = 0;
+    const readRates = [];
     for (const record of currentRecords()) {
         const cached = (record.cacheCreationTokens ?? 0) + (record.cacheReadTokens ?? 0);
         cacheCreate += record.cacheCreationTokens ?? 0;
@@ -380,11 +419,16 @@ export function getCacheStats() {
         }
         meteredCached += cached;
         // Savings = what cache-read tokens would have cost at full input rate,
-        // minus what they actually cost at 0.1×. (Cache creation is a slight
-        // *penalty* of 0.25× — netted in for honest reporting.)
+        // minus what they cost at the model's own read rate. This hardcoded 0.9
+        // (a 0.1 read) for every provider, so Kimi and Qwen (0.2) over-reported
+        // savings while DeepSeek (0.02) and Fable 5.1 (0.025) under-reported them.
+        // (Cache creation is a slight *penalty* of 0.25× — netted in.)
         const pricing = MODEL_PRICING[record.model];
         if (pricing) {
-            const cReadSaved = ((record.cacheReadTokens ?? 0) / 1_000_000) * pricing.inputPer1M * 0.9;
+            const readRate = cacheReadRateFor(record.model, record.provider);
+            if ((record.cacheReadTokens ?? 0) > 0)
+                readRates.push(readRate);
+            const cReadSaved = ((record.cacheReadTokens ?? 0) / 1_000_000) * pricing.inputPer1M * (1 - readRate);
             const cCreateCost = ((record.cacheCreationTokens ?? 0) / 1_000_000) * pricing.inputPer1M * 0.25;
             savings += cReadSaved - cCreateCost;
         }
@@ -395,6 +439,7 @@ export function getCacheStats() {
         estimatedSavingsUsd: Math.max(0, savings),
         hasFlatFeeCacheUsage: flatFeeCached > 0,
         isEntirelyFlatFeeCache: flatFeeCached > 0 && meteredCached === 0,
+        cacheReadRates: readRates,
     };
 }
 /**
@@ -506,7 +551,7 @@ export function formatCostReport() {
         // The billing multipliers only describe a metered account. On a plan
         // nothing is billed per token, so quoting a rate there would be as invented
         // as the per-model prices this report already refuses to show.
-        const readNote = cache.isEntirelyFlatFeeCache ? '' : ' (billed at 0.1× input rate)';
+        const readNote = cache.isEntirelyFlatFeeCache ? '' : formatCacheReadRates(cache.cacheReadRates);
         const writeNote = cache.isEntirelyFlatFeeCache ? '' : ' (billed at 1.25× input rate)';
         lines.push(`**Cache reads:** ${formatTokenCount(cache.cacheReadTokens)} tokens${readNote}`);
         if (cache.cacheCreationTokens > 0) {

package/dist/version.d.ts

@@ -1 +1 @@
-export declare const VERSION = "3.3.0";
+export declare const VERSION = "3.3.2";

package/dist/version.js

@@ -1,4 +1,4 @@
 // AUTO-GENERATED by scripts/gen-version.js — do not edit by hand.
 // Baked from package.json at build time so the bun-compiled binary reports
 // the right version (it has no package.json on disk to read at runtime).
-export const VERSION = '3.3.0';
+export const VERSION = '3.3.2';

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "codeep",
-  "version": "3.3.0",
+  "version": "3.3.2",
   "description": "AI-powered coding assistant built for the terminal. Multiple LLM providers, project-aware context, and a seamless development workflow.",
   "type": "module",
   "main": "dist/index.js",
@@ -17,6 +17,7 @@
     "test:coverage": "vitest run --coverage",
     "version": "node scripts/gen-version.js && git add src/version.ts",
     "export:catalogue": "node --import tsx scripts/export-catalogue.ts",
+    "export:commands": "node --import tsx scripts/export-commands.ts",
     "release": "node scripts/release.js"
   },
   "repository": {