@leo000001/opencode-quota-sidebar 2.0.1 → 2.0.4

package/CHANGELOG.md

@@ -2,6 +2,9 @@
 
 ## Unreleased
 
+- Add built-in `kimi-for-coding` subscription quota support via `GET https://api.kimi.com/coding/v1/usages`.
+- Parse Kimi's `5h` and `Weekly` windows, including reset timestamps, and render them like other subscription providers.
+- Accept OpenCode provider discovery responses that expose Kimi API keys through provider `key` fields.
 - Add Buzz API balance support for OpenAI-compatible providers that use a Buzz `baseURL`.
 - Document Buzz configuration, rendering, and outbound billing endpoints.
 - Keep session measured cost aligned with OpenCode root-session `message.cost` while still including descendant subagent usage in API-equivalent cost.

package/CONTRIBUTING.md

@@ -19,6 +19,7 @@ The plugin now uses a provider adapter registry, so adding a new provider does n
 - `baseURL` match: best for OpenAI-compatible relays such as RightCode or Buzz
 - Prefix/variant normalization: best when one provider has multiple runtime IDs
 - Balance-only providers should prefer `balance` over inventing fake percent windows
+- Built-in API-key providers such as `kimi-for-coding` may need both: direct ID matching for the canonical provider and support for OpenCode's discovered `key -> options.apiKey` bridge
 
 ## Add a new provider
 
@@ -70,6 +71,10 @@ If your provider is an OpenAI-compatible relay, prefer matching on
 `providerOptions.baseURL` instead of the runtime `providerID`; that keeps custom
 aliases working without extra user config.
 
+If your provider is built into OpenCode and already has a stable runtime ID
+(for example `kimi-for-coding`), prefer a direct provider-ID match first, then
+add a `baseURL` fallback only when it helps older/custom runtime shapes.
+
 If the new provider should appear in default `quota_summary` reports even when
 it has not yet been used in the current session, also update
 `listDefaultQuotaProviderIDs()` in `src/quota.ts`.
@@ -98,7 +103,7 @@ At minimum:
 - format output if using special fields (e.g. `balance`)
 - cache compatibility if the change replaces an older snapshot shape
 - mixed-provider rendering if the new provider will commonly appear next to
-  OpenAI/Copilot/RightCode in sidebar or toast output
+  OpenAI/Copilot/Kimi/RightCode in sidebar or toast output
 
 If the provider introduces new rendering rules or multi-window behavior, add
 coverage in both `src/__tests__/quota.test.ts` and `src/__tests__/format.test.ts`.

package/README.md

@@ -13,7 +13,7 @@ Add the package name to `plugin` in your `opencode.json`. OpenCode uses Bun to i
 
 ```json
 {
-  "plugin": ["@leo000001/opencode-quota-sidebar@2.0.0"]
+  "plugin": ["@leo000001/opencode-quota-sidebar@2.0.1"]
 }
 ```
 
@@ -41,13 +41,14 @@ On Windows, use forward slashes: `"file:///D:/Lab/opencode-quota-sidebar/dist/in
 
 ## Supported quota providers
 
-| Provider       | Endpoint                               | Auth            | Status                                  |
-| -------------- | -------------------------------------- | --------------- | --------------------------------------- |
-| OpenAI Codex   | `chatgpt.com/backend-api/wham/usage`   | OAuth (ChatGPT) | Multi-window (short-term + weekly)      |
-| GitHub Copilot | `api.github.com/copilot_internal/user` | OAuth           | Monthly quota                           |
-| RightCode      | `www.right.codes/account/summary`      | API key         | Subscription or balance (by prefix)     |
-| Buzz           | `buzzai.cc/v1/dashboard/billing/*`     | API key         | Balance only (computed from total-used) |
-| Anthropic      | `api.anthropic.com/api/oauth/usage`    | OAuth           | Multi-window (5h + weekly / plan-based) |
+| Provider       | Endpoint                               | Auth            | Status                                  |
+| -------------- | -------------------------------------- | --------------- | --------------------------------------- |
+| OpenAI Codex   | `chatgpt.com/backend-api/wham/usage`   | OAuth (ChatGPT) | Multi-window (short-term + weekly)      |
+| GitHub Copilot | `api.github.com/copilot_internal/user` | OAuth           | Monthly quota                           |
+| Kimi For Coding | `api.kimi.com/coding/v1/usages`       | API key         | Multi-window subscription (5h + weekly) |
+| RightCode      | `www.right.codes/account/summary`      | API key         | Subscription or balance (by prefix)     |
+| Buzz           | `buzzai.cc/v1/dashboard/billing/*`     | API key         | Balance only (computed from total-used) |
+| Anthropic      | `api.anthropic.com/api/oauth/usage`    | OAuth           | Multi-window (5h + weekly / plan-based) |
 
 Want to add support for another provider (Google Antigravity, Zhipu AI, Firmware AI, etc.)? See [CONTRIBUTING.md](CONTRIBUTING.md).
 
@@ -64,23 +65,32 @@ Want to add support for another provider (Google Antigravity, Zhipu AI, Firmware
   - quota lines: quota text like `OpenAI 5h 80% Rst 16:20`; short windows (`5h`, `1d`, `Daily`) show `HH:MM` on same-day resets and `MM-DD HH:MM` when crossing days, while longer windows continue to show `MM-DD`
   - RightCode daily quota shows `$remaining/$dailyTotal` + expiry (e.g. `RC Daily $105/$60 Exp 02-27`, without trailing percent) and also shows balance on the next indented line when available; `Exp` remains date-only
 - Session-scoped usage/quota can include descendant subagent sessions (enabled by default via `sidebar.includeChildren=true`). Traversal is bounded by `childrenMaxDepth` (default 6), `childrenMaxSessions` (default 128), and `childrenConcurrency` (default 5); truncation is logged when `OPENCODE_QUOTA_DEBUG=1`. Day/week/month ranges never merge children — only session scope does.
-- Toast message includes three sections: `Token Usage`, `Cost as API` (per provider), and `Quota`
+- Toast message can include four sections: `Token Usage`, `Cost as API` (per provider), `Provider Cache` (when provider-level cache coverage is available), and `Quota`
 - `quota_summary` markdown / toast also include `Cache Coverage` and `Cache Read Coverage` summary lines when available
 - Quota snapshots are de-duplicated before rendering to avoid repeated provider lines
 - Custom tools:
   - `quota_summary` — generate usage report for session/day/week/month (markdown + toast)
 - `quota_show` — toggle sidebar title display on/off (state persists across sessions)
-- After startup, titles refresh on the next relevant session/message event or when `quota_show` is toggled
-- Quota connectors:
-  - OpenAI Codex OAuth (`/backend-api/wham/usage`)
-  - GitHub Copilot OAuth (`/copilot_internal/user`)
-  - RightCode API key (`/account/summary`)
-  - Buzz API key (`/v1/dashboard/billing/subscription` + `/v1/dashboard/billing/usage`)
-  - Anthropic Claude OAuth (`/api/oauth/usage`, with beta header)
+- After startup, titles are restored immediately when persisted display mode is OFF; when persisted display mode is ON, touched titles refresh on startup and the rest update on the next relevant session/message event or when `quota_show` is toggled
+- Quota connectors:
+  - OpenAI Codex OAuth (`/backend-api/wham/usage`)
+  - GitHub Copilot OAuth (`/copilot_internal/user`)
+  - Kimi For Coding API key (`/usages`, built-in `kimi-for-coding` provider)
+  - RightCode API key (`/account/summary`)
+  - Buzz API key (`/v1/dashboard/billing/subscription` + `/v1/dashboard/billing/usage`)
+  - Anthropic Claude OAuth (`/api/oauth/usage`, with beta header)
 - OpenAI OAuth quota checks auto-refresh expired access token (using refresh token)
-- API key providers still show usage aggregation (quota only applies to subscription providers)
-- Incremental usage aggregation — only processes new messages since last cursor
-- Sidebar token units are adaptive (`k`/`m` with one decimal where applicable)
+- API key providers still show usage aggregation (quota only applies to subscription providers)
+- Incremental usage aggregation — only processes new messages since last cursor
+- Sidebar token units are adaptive (`k`/`m` with one decimal where applicable)
+
+### Kimi For Coding notes
+
+- OpenCode's built-in provider ID is `kimi-for-coding` and its runtime base URL is `https://api.kimi.com/coding/v1`.
+- The plugin treats Kimi as a subscription quota source, not a balance source.
+- Quota data is read from `GET https://api.kimi.com/coding/v1/usages`.
+- The current implementation maps the short rolling window in `limits[]` to `5h` and the top-level `usage` block to `Weekly`.
+- Rendering follows the same compact reset formatting as OpenAI: short windows show `Rst MM-DD HH:MM` when they cross days, and longer windows show `Rst MM-DD`.
 
 ## Storage layout
 
@@ -90,12 +100,17 @@ The plugin stores lightweight global state and date-partitioned session chunks.
   - `titleEnabled`
   - `sessionDateMap` (sessionID -> `YYYY-MM-DD`)
   - `quotaCache`
-- Session chunks: `<opencode-data>/quota-sidebar-sessions/YYYY/MM/DD.json`
-  - per-session title state (`baseTitle`, `lastAppliedTitle`)
-  - `createdAt`
-  - `parentID` (when the session is a subagent child session)
-  - cached usage summary used by `quota_summary`
-  - incremental aggregation cursor
+- Session chunks: `<opencode-data>/quota-sidebar-sessions/YYYY/MM/DD.json`
+  - per-session title state (`baseTitle`, `lastAppliedTitle`)
+  - `createdAt`
+  - `parentID` (when the session is a subagent child session)
+  - cached usage summary used by `quota_summary`, including session-level and provider-level `cacheBuckets` for cache coverage reporting
+  - incremental aggregation cursor
+
+Notes on cache coverage persistence:
+
+- Older cached usage written before `cacheBuckets` existed can only be approximated from top-level `cache_read` / `cache_write` totals.
+- In those legacy cases, mixed read-only + read-write cache traffic may be attributed to a single fallback bucket until the session is recomputed from messages.
 
 Example tree:
 
@@ -450,12 +465,13 @@ Set `OPENCODE_QUOTA_DEBUG=1` to enable debug logging to stderr. This logs:
 ## Security & privacy notes
 
 - The plugin reads OpenCode credentials from `<opencode-data>/auth.json`.
-- If enabled, quota checks call external endpoints:
-  - OpenAI Codex: `https://chatgpt.com/backend-api/wham/usage`
-  - GitHub Copilot: `https://api.github.com/copilot_internal/user`
-  - RightCode: `https://www.right.codes/account/summary`
-  - Buzz: `https://buzzai.cc/v1/dashboard/billing/subscription` and `https://buzzai.cc/v1/dashboard/billing/usage`
-  - Anthropic: `https://api.anthropic.com/api/oauth/usage`
+- If enabled, quota checks call external endpoints:
+  - OpenAI Codex: `https://chatgpt.com/backend-api/wham/usage`
+  - GitHub Copilot: `https://api.github.com/copilot_internal/user`
+  - Kimi For Coding: `https://api.kimi.com/coding/v1/usages`
+  - RightCode: `https://www.right.codes/account/summary`
+  - Buzz: `https://buzzai.cc/v1/dashboard/billing/subscription` and `https://buzzai.cc/v1/dashboard/billing/usage`
+  - Anthropic: `https://api.anthropic.com/api/oauth/usage`
 - **Screen-sharing warning**: Session titles and toasts surface usage/quota
   information. If you are screen-sharing or recording, consider toggling the
   sidebar display off (`/qtoggle` or `quota_show` tool) to avoid leaking
@@ -465,8 +481,9 @@ Set `OPENCODE_QUOTA_DEBUG=1` to enable debug logging to stderr. This logs:
 - OpenAI OAuth token refresh is disabled by default; set
   `quota.refreshAccessToken=true` if you want the plugin to refresh access
   tokens when expired.
-- Anthropic quota currently uses a beta/internal-style OAuth usage endpoint and
-  request header; response fields may change without notice.
+- Anthropic quota currently uses a beta/internal-style OAuth usage endpoint and
+  request header; response fields may change without notice.
+- Kimi For Coding quota uses the current `/usages` response shape exposed by the Kimi coding service; if Kimi changes that payload, window parsing may need to be updated.
 - State/chunk file writes refuse to write through symlinked targets (best-effort defense-in-depth).
 - The `OPENCODE_QUOTA_DATA_HOME` env var overrides the OpenCode data directory
   path (for testing); do not set this in production.

package/SECURITY.md

@@ -25,7 +25,7 @@ We will acknowledge reports as quickly as possible and provide a remediation tim
 - Keep debug logs free of secrets.
 - Prefer fail-closed behavior for writes (already enforced via symlink checks and atomic writes).
 - Quota fetching may contact provider-operated endpoints such as OpenAI, GitHub,
-  RightCode, Buzz, and Anthropic; review any new provider integration for
+  Kimi, RightCode, Buzz, and Anthropic; review any new provider integration for
   outbound data exposure and header/token handling.
 - Some quota integrations rely on beta or internal-style endpoints; document
   instability risks clearly and avoid assuming long-term API compatibility.

package/dist/format.js

@@ -1,4 +1,4 @@
-import { getCacheCoverageMetrics } from './usage.js';
+import { getCacheCoverageMetrics, getProviderCacheCoverageMetrics, } from './usage.js';
 import { canonicalProviderID, collapseQuotaSnapshots, displayShortLabel, quotaDisplayLabel, } from './quota_render.js';
 import { stripAnsi } from './title.js';
 /** M6 fix: handle negative, NaN, Infinity gracefully. */
@@ -161,6 +161,16 @@ function formatPercent(value, decimals = 1) {
     const pct = (safe * 100).toFixed(decimals);
     return `${pct.replace(/\.0+$/, '').replace(/(\.\d*[1-9])0+$/, '$1')}%`;
 }
+function formatQuotaPercent(value, options) {
+    const missing = options?.missing ?? '-';
+    if (value === undefined)
+        return missing;
+    if (!Number.isFinite(value) || value < 0)
+        return missing;
+    if (options?.rounded)
+        return `${Math.round(value)}%`;
+    return `${value.toFixed(options?.decimals ?? 1)}%`;
+}
 function alignPairs(pairs, indent = '  ') {
     if (pairs.length === 0)
         return [];
@@ -187,9 +197,10 @@ function compactQuotaInline(quota) {
         const first = quota.windows[0];
         const showPercent = first.showPercent !== false;
         const firstLabel = sanitizeLine(first.label || '');
-        const pct = first.remainingPercent === undefined
-            ? undefined
-            : `${Math.round(first.remainingPercent)}%`;
+        const pct = formatQuotaPercent(first.remainingPercent, {
+            rounded: true,
+            missing: '',
+        });
         const summary = showPercent
             ? [firstLabel, pct].filter(Boolean).join(' ')
             : firstLabel.replace(/^Daily\s+/i, '') || firstLabel;
@@ -200,8 +211,12 @@ function compactQuotaInline(quota) {
     if (quota.balance) {
         return `${label} Balance ${formatCurrency(quota.balance.amount, quota.balance.currency)}`;
     }
-    if (quota.remainingPercent !== undefined) {
-        return `${label} ${Math.round(quota.remainingPercent)}%`;
+    const singlePercent = formatQuotaPercent(quota.remainingPercent, {
+        rounded: true,
+        missing: '',
+    });
+    if (singlePercent) {
+        return `${label} ${singlePercent}`;
     }
     return label;
 }
@@ -351,9 +366,7 @@ function compactQuotaWide(quota, labelWidth = 0, options) {
         : undefined;
     const renderWindow = (win) => {
         const showPercent = win.showPercent !== false;
-        const pct = win.remainingPercent === undefined
-            ? '?'
-            : `${Math.round(win.remainingPercent)}%`;
+        const pct = formatQuotaPercent(win.remainingPercent, { rounded: true });
         const parts = win.label
             ? showPercent
                 ? [sanitizeLine(win.label), pct]
@@ -387,9 +400,7 @@ function compactQuotaWide(quota, labelWidth = 0, options) {
         return maybeBreak(balanceText, [balanceText]);
     }
     // Fallback: single value from top-level remainingPercent
-    const percent = quota.remainingPercent === undefined
-        ? '?'
-        : `${Math.round(quota.remainingPercent)}%`;
+    const percent = formatQuotaPercent(quota.remainingPercent, { rounded: true });
     const reset = compactReset(quota.resetAt, 'Rst');
     const fallbackText = `Remaining ${percent}${reset ? ` Rst ${reset}` : ''}`;
     return maybeBreak(fallbackText, [fallbackText]);
@@ -499,12 +510,71 @@ export function renderMarkdownReport(period, usage, quotas, options) {
             return '-';
         return formatApiCostValue(usage.apiCost);
     };
-    const providerRows = Object.values(usage.providers)
-        .sort((a, b) => b.total - a.total)
+    const cacheCoverageCell = (provider) => {
+        const metrics = getProviderCacheCoverageMetrics(provider);
+        return metrics.cacheCoverage !== undefined
+            ? formatPercent(metrics.cacheCoverage, 1)
+            : '-';
+    };
+    const cacheReadCoverageCell = (provider) => {
+        const metrics = getProviderCacheCoverageMetrics(provider);
+        return metrics.cacheReadCoverage !== undefined
+            ? formatPercent(metrics.cacheReadCoverage, 1)
+            : '-';
+    };
+    const providerEntries = Object.values(usage.providers).sort((a, b) => b.total - a.total);
+    const highlightLines = () => {
+        const lines = [];
+        const providerLabel = (providerID) => quotaDisplayLabel({
+            providerID,
+            label: providerID,
+            status: 'ok',
+            checkedAt: 0,
+        });
+        const topApiCost = providerEntries
+            .filter((provider) => provider.apiCost > 0)
+            .sort((a, b) => b.apiCost - a.apiCost)[0];
+        if (topApiCost) {
+            lines.push(`- Top API cost: ${quotaDisplayLabel({
+                providerID: topApiCost.providerID,
+                label: topApiCost.providerID,
+                status: 'ok',
+                checkedAt: 0,
+            })} (${formatUsd(topApiCost.apiCost)})`);
+        }
+        const bestCacheCoverage = providerEntries
+            .map((provider) => ({
+            provider,
+            value: getProviderCacheCoverageMetrics(provider).cacheCoverage,
+        }))
+            .filter((entry) => entry.value !== undefined)
+            .sort((a, b) => b.value - a.value)[0];
+        if (bestCacheCoverage) {
+            lines.push(`- Best Cache Coverage: ${providerLabel(bestCacheCoverage.provider.providerID)} (${formatPercent(bestCacheCoverage.value, 1)})`);
+        }
+        const bestCacheReadCoverage = providerEntries
+            .map((provider) => ({
+            provider,
+            value: getProviderCacheCoverageMetrics(provider).cacheReadCoverage,
+        }))
+            .filter((entry) => entry.value !== undefined)
+            .sort((a, b) => b.value - a.value)[0];
+        if (bestCacheReadCoverage) {
+            lines.push(`- Best Cache Read Coverage: ${providerLabel(bestCacheReadCoverage.provider.providerID)} (${formatPercent(bestCacheReadCoverage.value, 1)})`);
+        }
+        const highestMeasured = providerEntries
+            .filter((provider) => measuredCostCell(provider.providerID, provider.cost) !== '-')
+            .sort((a, b) => b.cost - a.cost)[0];
+        if (highestMeasured && highestMeasured.cost > 0) {
+            lines.push(`- Highest measured cost: ${providerLabel(highestMeasured.providerID)} (${formatUsd(highestMeasured.cost)})`);
+        }
+        return lines;
+    };
+    const providerRows = providerEntries
         .map((provider) => {
         const providerID = mdCell(provider.providerID);
         return showCost
-            ? `| ${providerID} | ${shortNumber(provider.input)} | ${shortNumber(provider.output)} | ${shortNumber(provider.cacheRead + provider.cacheWrite)} | ${shortNumber(provider.total)} | ${measuredCostCell(provider.providerID, provider.cost)} | ${apiCostCell(provider.providerID, provider.apiCost)} |`
+            ? `| ${providerID} | ${shortNumber(provider.input)} | ${shortNumber(provider.output)} | ${shortNumber(provider.cacheRead + provider.cacheWrite)} | ${shortNumber(provider.total)} | ${cacheCoverageCell(provider)} | ${cacheReadCoverageCell(provider)} | ${measuredCostCell(provider.providerID, provider.cost)} | ${apiCostCell(provider.providerID, provider.apiCost)} |`
             : `| ${providerID} | ${shortNumber(provider.input)} | ${shortNumber(provider.output)} | ${shortNumber(provider.cacheRead + provider.cacheWrite)} | ${shortNumber(provider.total)} |`;
     });
     const quotaLines = collapseQuotaSnapshots(quotas).flatMap((quota) => {
@@ -516,9 +586,7 @@ export function renderMarkdownReport(period, usage, quotas, options) {
                     const winLabel = win.label ? ` (${win.label})` : '';
                     return mdCell(`- ${displayLabel}${winLabel}: ${quota.status} | reset ${reportResetLine(win.resetAt, win.resetLabel, win.label)}`);
                 }
-                const remaining = win.remainingPercent === undefined
-                    ? '-'
-                    : `${win.remainingPercent.toFixed(1)}%`;
+                const remaining = formatQuotaPercent(win.remainingPercent);
                 const winLabel = win.label ? ` (${win.label})` : '';
                 return mdCell(`- ${displayLabel}${winLabel}: ${quota.status} | remaining ${remaining} | reset ${reportResetLine(win.resetAt, win.resetLabel, win.label)}`);
             });
@@ -537,9 +605,7 @@ export function renderMarkdownReport(period, usage, quotas, options) {
                 mdCell(`- ${displayLabel}: ${quota.status}${quota.note ? ` | ${quota.note}` : ''}`),
             ];
         }
-        const remaining = quota.remainingPercent === undefined
-            ? '-'
-            : `${quota.remainingPercent.toFixed(1)}%`;
+        const remaining = formatQuotaPercent(quota.remainingPercent);
         return [
             mdCell(`- ${displayLabel}: ${quota.status} | remaining ${remaining} | reset ${reportResetLine(quota.resetAt)}${quota.note ? ` | ${quota.note}` : ''}`),
         ];
@@ -564,17 +630,20 @@ export function renderMarkdownReport(period, usage, quotas, options) {
                 `- API cost: ${apiCostSummaryValue()}`,
             ]
             : []),
+        ...(highlightLines().length > 0
+            ? ['', '### Highlights', ...highlightLines()]
+            : []),
         '',
         '### Usage by Provider',
         showCost
-            ? '| Provider | Input | Output | Cache | Total | Measured Cost | API Cost |'
+            ? '| Provider | Input | Output | Cache | Total | Cache Coverage | Cache Read Coverage | Measured Cost | API Cost |'
             : '| Provider | Input | Output | Cache | Total |',
         showCost
-            ? '|---|---:|---:|---:|---:|---:|---:|'
+            ? '|---|---:|---:|---:|---:|---:|---:|---:|---:|'
             : '|---|---:|---:|---:|---:|',
         ...(providerRows.length
             ? providerRows
-            : [showCost ? '| - | - | - | - | - | - | - |' : '| - | - | - | - | - |']),
+            : [showCost ? '| - | - | - | - | - | - | - | - | - |' : '| - | - | - | - | - |']),
         '',
         '### Subscription Quota',
         ...(quotaLines.length
@@ -638,14 +707,35 @@ export function renderToastMessage(period, usage, quotas, options) {
             lines.push(fitLine(hasAnyUsage ? '  N/A (Copilot)' : '  -', width));
         }
     }
+    const providerCachePairs = Object.values(usage.providers)
+        .map((provider) => {
+        const metrics = getProviderCacheCoverageMetrics(provider);
+        const parts = [];
+        if (metrics.cacheCoverage !== undefined) {
+            parts.push(`Cov ${formatPercent(metrics.cacheCoverage, 1)}`);
+        }
+        if (metrics.cacheReadCoverage !== undefined) {
+            parts.push(`Read ${formatPercent(metrics.cacheReadCoverage, 1)}`);
+        }
+        if (parts.length === 0)
+            return undefined;
+        return {
+            label: displayShortLabel(provider.providerID),
+            value: parts.join('  '),
+        };
+    })
+        .filter((item) => Boolean(item));
+    if (providerCachePairs.length > 0) {
+        lines.push('');
+        lines.push(fitLine('Provider Cache', width));
+        lines.push(...alignPairs(providerCachePairs).map((line) => fitLine(line, width)));
+    }
     const quotaPairs = collapseQuotaSnapshots(quotas).flatMap((item) => {
         if (item.status === 'ok') {
             if (item.windows && item.windows.length > 0) {
                 const pairs = item.windows.map((win, idx) => {
                     const showPercent = win.showPercent !== false;
-                    const pct = win.remainingPercent === undefined
-                        ? '-'
-                        : `${win.remainingPercent.toFixed(1)}%`;
+                    const pct = formatQuotaPercent(win.remainingPercent);
                     const reset = compactReset(win.resetAt, win.resetLabel, win.label);
                     const parts = [win.label];
                     if (showPercent)
@@ -673,9 +763,7 @@ export function renderToastMessage(period, usage, quotas, options) {
                     },
                 ];
             }
-            const percent = item.remainingPercent === undefined
-                ? '-'
-                : `${item.remainingPercent.toFixed(1)}%`;
+            const percent = formatQuotaPercent(item.remainingPercent);
             const reset = compactReset(item.resetAt, 'Rst');
             return [
                 {

package/dist/index.js

@@ -12,6 +12,8 @@ import { createPersistenceScheduler } from './persistence.js';
 import { createQuotaService } from './quota_service.js';
 import { createUsageService } from './usage_service.js';
 import { createTitleApplicator } from './title_apply.js';
+const SHUTDOWN_HOOK_KEY = Symbol.for('opencode-quota-sidebar.shutdown-hook');
+const SHUTDOWN_CALLBACKS_KEY = Symbol.for('opencode-quota-sidebar.shutdown-callbacks');
 export async function QuotaSidebarPlugin(input) {
     const quotaRuntime = createQuotaRuntime();
     const configDir = resolveOpencodeConfigDir();
@@ -164,18 +166,61 @@ export async function QuotaSidebarPlugin(input) {
         restoreConcurrency: RESTORE_TITLE_CONCURRENCY,
     });
     const titleRefresh = createTitleRefreshScheduler({
-        apply: titleApplicator.applyTitle,
+        apply: async (sessionID) => {
+            await titleApplicator.applyTitle(sessionID);
+        },
         onError: swallow('titleRefresh'),
     });
     scheduleTitleRefresh = titleRefresh.schedule;
     const restoreAllVisibleTitles = titleApplicator.restoreAllVisibleTitles;
     const refreshAllTouchedTitles = titleApplicator.refreshAllTouchedTitles;
     const refreshAllVisibleTitles = titleApplicator.refreshAllVisibleTitles;
+    let startupTitleWork = Promise.resolve();
+    const runStartupRestore = async (attempt = 0) => {
+        const result = await restoreAllVisibleTitles({
+            abortIfEnabled: config.sidebar.enabled,
+        });
+        if (result.restored === result.attempted)
+            return;
+        debug(`startup restore incomplete: restored ${result.restored}/${result.attempted} touched titles while display mode remains OFF`);
+        if (state.titleEnabled || config.sidebar.enabled === false)
+            return;
+        if (attempt >= 2)
+            return;
+        await new Promise((resolve) => setTimeout(resolve, 1_000));
+        await runStartupRestore(attempt + 1);
+    };
     if (!state.titleEnabled || !config.sidebar.enabled) {
-        void restoreAllVisibleTitles().catch(swallow('startup:restoreAllVisibleTitles'));
+        startupTitleWork = runStartupRestore().catch(swallow('startup:restoreAllVisibleTitles'));
     }
     else {
-        void refreshAllTouchedTitles().catch(swallow('startup:refreshAllTouchedTitles'));
+        startupTitleWork = refreshAllTouchedTitles()
+            .then(() => undefined)
+            .catch(swallow('startup:refreshAllTouchedTitles'));
+    }
+    const shutdown = async () => {
+        await Promise.race([
+            startupTitleWork,
+            new Promise((resolve) => setTimeout(resolve, 5_000)),
+        ]).catch(swallow('shutdown:startupTitleWork'));
+        await titleRefresh.waitForQuiescence().catch(swallow('shutdown:titleQuiescence'));
+        await flushSave().catch(swallow('shutdown:flushSave'));
+    };
+    const processWithHook = process;
+    const shutdownCallbacks = (processWithHook[SHUTDOWN_CALLBACKS_KEY] ||= new Set());
+    shutdownCallbacks.add(shutdown);
+    if (!processWithHook[SHUTDOWN_HOOK_KEY]) {
+        processWithHook[SHUTDOWN_HOOK_KEY] = true;
+        process.once('beforeExit', () => {
+            void Promise.allSettled(Array.from(shutdownCallbacks).map((callback) => callback()));
+        });
+        for (const signal of ['SIGINT', 'SIGTERM']) {
+            process.once(signal, () => {
+                void Promise.allSettled(Array.from(shutdownCallbacks).map((callback) => callback())).finally(() => {
+                    process.kill(process.pid, signal);
+                });
+            });
+        }
     }
     const showToast = async (period, message) => {
         await input.client.tui
@@ -268,9 +313,12 @@ export async function QuotaSidebarPlugin(input) {
             },
             scheduleSave,
             flushSave,
+            waitForStartupTitleWork: () => startupTitleWork,
             refreshSessionTitle: (sessionID, delay) => titleRefresh.schedule(sessionID, delay ?? 250),
             cancelAllTitleRefreshes: () => titleRefresh.cancelAll(),
+            flushScheduledTitleRefreshes: () => titleRefresh.flushScheduled(),
             waitForTitleRefreshIdle: () => titleRefresh.waitForIdle(),
+            waitForTitleRefreshQuiescence: () => titleRefresh.waitForQuiescence(),
             restoreAllVisibleTitles,
             refreshAllTouchedTitles,
             refreshAllVisibleTitles,
@@ -279,7 +327,10 @@ export async function QuotaSidebarPlugin(input) {
             getQuotaSnapshots,
             renderMarkdownReport,
             renderToastMessage,
-            config,
+            config: {
+                sidebar: config.sidebar,
+                sidebarEnabled: config.sidebar.enabled,
+            },
         }),
     };
 }

package/dist/persistence.js

@@ -4,6 +4,17 @@ export function createPersistenceScheduler(deps) {
     let stateDirty = false;
     let saveTimer;
     let saveInFlight = Promise.resolve();
+    const scheduleRetry = (delayMs = 1_000) => {
+        if (saveTimer)
+            return;
+        saveTimer = setTimeout(() => {
+            saveTimer = undefined;
+            void persist().catch((error) => {
+                debug(`persistState retry failed: ${String(error)}`);
+                scheduleRetry(Math.min(delayMs * 2, 10_000));
+            });
+        }, delayMs);
+    };
     /**
      * Capture and delete specific dirty keys instead of clearing the whole set.
      * Keys added between capture and write completion are preserved.
@@ -34,7 +45,10 @@ export function createPersistenceScheduler(deps) {
             clearTimeout(saveTimer);
         saveTimer = setTimeout(() => {
             saveTimer = undefined;
-            void persist().catch(swallow('persistState:save'));
+            void persist().catch((error) => {
+                swallow('persistState:save')(error);
+                scheduleRetry();
+            });
         }, 200);
     };
     const flushSave = async () => {

package/dist/providers/core/kimi_for_coding.d.ts

@@ -0,0 +1,2 @@
+import type { QuotaProviderAdapter } from '../types.js';
+export declare const kimiForCodingAdapter: QuotaProviderAdapter;