Drop COMMANDCODE_DEBUG, COMMANDCODE_DEBUG_FILE, and COMMANDCODE_ZDR env vars

AGENTS.md

@@ -45,7 +45,7 @@ src/transform.ts  LanguageModelV3CallOptions -> /alpha/generate envelope (return
 src/events.ts     Async iterator over the NDJSON/SSE response body.
 src/usage.ts      finish event -> LanguageModelV3Usage; finish-reason unification.
 src/redact.ts     Credential scrubbing for error surfaces.
-src/log.ts        Opt-in tracing (COMMANDCODE_DEBUG) to a log file.
+src/log.ts        Opt-in tracing (toggle file) to a log file.
 src/toggles.ts    Shared toggle file (~/.config/opencode/commandcode-toggles.json).
 src/quota.ts      Live 5-hour/weekly/monthly quota from the CommandCode alpha billing API.
 src/tui.tsx       opencode TUI plugin: /cc-zdr, /cc-debug, /cc-status, /cc-usage + sidebar.
@@ -130,9 +130,14 @@ These are load-bearing. Breaking one causes silent failures in opencode.
     `provider.commandcode.models` in `opencode.json` and never asks a custom `file://` provider to
     discover models (only internal providers can register `discoverModels`). Refresh the map with
     `npm run sync-models`; do not expect a discovery hook in `src/` to populate it.
-13. **The TUI plugin lives outside the provider entrypoint.** `src/tui.tsx` is loaded from
+ 13. **The TUI plugin lives outside the provider entrypoint.** `src/tui.tsx` is loaded from
-    **source** (not `dist/`) via `opencode plugin <path>`, which writes a `tui.json` `plugin`
+    **source** (not `dist/`) via a `tui.json` `plugin` entry that must be a **direct `file://` URL
-    entry; it is exposed as `./tui` in `package.json`. opencode compiles the `.tsx` at load with
+    to `src/tui.tsx`** (global `~/.config/opencode/tui.json` or project `.opencode/tui.json`),
+    wired by hand: `opencode plugin <path>` cannot install a bare `.tsx` (it needs a `package.json`
+    manifest next to the target), and the spec must never point at the **repo root** — a directory
+    spec is also picked up as a server plugin via `package.json` `main` → `dist/index.js` (the
+    provider, not a plugin). Do not re-add a `./tui` export for this; it is deliberately removed.
+    opencode compiles the `.tsx` at load with
     its Bun/Solid transform and maps `solid-js` / `@opentui/solid` to its internal modules, so the
     package stays dependency-free — do not add a build step or real imports of those packages.
     `src/tui-shims.d.ts` is the only local stand-in for types; keep it in sync with what the file
@@ -154,9 +159,9 @@ These are load-bearing. Breaking one causes silent failures in opencode.
     the same `namespace: "palette"` commands the palette lists (both filter out `hidden: true`).
     A slash-only entry is not expressible; registering `/cc-*` also adds them to `Ctrl+P`.
 15. **Toggles are resolved per request, never at load.** `zdr` and `debug` are read in
-    `model.ts`/`log.ts` on every call from the toggle file / env, so `/cc-*` changes take effect
+    `model.ts`/`log.ts` on every call from the toggle file, so `/cc-*` changes take effect
     without restarting opencode. Precedence for `zdr`: `providerOptions.commandcode.zdr` >
-    `x-cmd-zdr` header > toggle file > `COMMANDCODE_ZDR`. Do not reintroduce load-time consts.
+    `x-cmd-zdr` header > toggle file. Do not reintroduce load-time consts.
 16. **Quota is live from the alpha billing API, not `server.py`.** `src/quota.ts` reads
     `/alpha/whoami`, `/alpha/billing/credits`, `/alpha/billing/subscriptions`, and
     `/alpha/usage/summary` (the same endpoints the `cmd` CLI `/usage` uses). `server.py` predates
@@ -165,6 +170,14 @@ These are load-bearing. Breaking one causes silent failures in opencode.
     plan; never hardcode a plan→cap table. The TUI resolves the key from `api.state` (provider
     entry, then `config.provider.commandcode.options`, then `COMMANDCODE_API_KEY`), expanding
     `{env:VAR}` itself, and must never log the key — quota errors go through `redact()`.
+17. **Retryable upstream failures can arrive inside an HTTP 200 stream.** CommandCode's gateway
+    answers `200` and then emits an SSE `error` event carrying `statusCode`/`isRetryable` (e.g.
+    `{"type":"server_error","message":"Invalid error response format: Gateway request failed",
+    "statusCode":520,"isRetryable":true}`). `fetchWithRetry` buffers only the `start`/`start-step`
+    preamble and, if the first real event is such a retryable error and attempts remain, cancels the
+    body and retries with backoff. Once any content event is seen the error is surfaced, not
+    retried (retrying mid-stream would duplicate output). Non-retryable in-stream errors (400s)
+    surface immediately. Keep the preamble set in sync if upstream adds new pre-content events.
 
 ## Change workflow
 

README.md

@@ -48,7 +48,7 @@ npm run build
   "provider": {
     "commandcode": {
       "name": "Command Code",
-      "npm": "file:///C:/DevTools/pienv/ccprovider/dist/index.js",
+      "npm": "file:///C:/dev/opencode-commandcode-provider/dist/index.js",
       "options": {
         "apiKey": "{env:COMMANDCODE_API_KEY}"
       },
@@ -99,6 +99,11 @@ without writing. Each row shows the model id and name plus a second line with it
 (`ctx 1M (1000000)`); the catalog carries no pricing fields, so no cost is shown. Needs an
 interactive terminal.
 
+Before anything else the script requires a CommandCode API key (from
+`provider.commandcode.options.apiKey` or `COMMANDCODE_API_KEY`). If neither is present — including
+when `opencode.json` does not exist yet — it exits and points you at the CommandCode Console
+(<https://commandcode.ai/studio/provider>) to generate one. It never writes without a key.
+
 Then restart opencode and confirm:
 
 ```powershell
@@ -141,22 +146,33 @@ global config (`baseURL`/`ccVersion` are read from there too).
 `options.headers` (with a leading `Bearer ` stripped). If neither is present, requests are sent
 unauthenticated and CommandCode will reject them.
 
+## Environment variables
+
+All environment variables read by the provider, its scripts, and its TUI plugin. String values in
+`options` also accept `{env:NAME}`, which opencode/the TUI expand generically (not a fixed set).
+
+CommandCode-specific:
+
+| Name | Purpose | Default |
+| --- | --- | --- |
+| `COMMANDCODE_API_KEY` | CommandCode API key for `smoke`/`quota`/`sync-models` and the TUI's key resolution; referenced from config as `{env:COMMANDCODE_API_KEY}`. | — |
+| `COMMANDCODE_TOGGLES_FILE` | Overrides the shared toggle-file path. | `$XDG_CONFIG_HOME/opencode/commandcode-toggles.json` |
+| `COMMANDCODE_QUOTA_INTERVAL_MS` | Sidebar fallback quota refresh interval in ms. | `300000` (floor `120000`) |
+
+Standard:
+
+| Name | Purpose | Default |
+| --- | --- | --- |
+| `XDG_CONFIG_HOME` | Base config dir for the toggle file. | `~/.config` |
+| `NO_COLOR` | Disables ANSI colors in the `sync-models` picker (any value). | unset |
+| `TERM` | `dumb` disables ANSI colors in the `sync-models` picker. | — |
+
 ## Debug tracing
 
-Set `COMMANDCODE_DEBUG=1` to write a trace of every request and stream event to a log file.
+Flip the `debug` key in the shared toggle file (via `/cc-debug`) to write a trace of every
-Silent (and no file is created) unless enabled. Every line is passed through `redact()` so
+request and stream event to a log file. Silent (and no file is created) unless enabled; the log
-credentials never reach disk. It can also be flipped at runtime with `/cc-debug` (see
+path is `<os tempdir>/commandcode-debug.log`. Every line is passed through `redact()` so
-[Runtime toggles](#runtime-toggles)).
+credentials never reach disk.
-
-```powershell
-$env:COMMANDCODE_DEBUG = "1"
-opencode run "Reply with exactly: pong" -m commandcode/deepseek/deepseek-v4.1-flash
-```
-
-| Env var | Default | Description |
-| --- | --- | --- |
-| `COMMANDCODE_DEBUG` | unset | `1`/`true`/`yes` enables tracing. |
-| `COMMANDCODE_DEBUG_FILE` | `<os tempdir>/commandcode-debug.log` | Where the trace is appended. |
 
 The log captures the request (model id, body byte length), each HTTP attempt (status, retry
 wait, elapsed ms), every stream event payload, and the terminal finish reason + usage. It is
@@ -165,8 +181,8 @@ useful for diagnosing model selection, retry, tool-call, and finish-reason issue
 ## ZDR header toggle
 
 The `x-cmd-zdr: 1` request header is **omitted by default** because it is rejected by some
-models. It can be enabled with the `COMMANDCODE_ZDR` environment variable, the `/cc-zdr` slash
+models. It can be enabled with the `/cc-zdr` slash command or
-command, or `providerOptions.commandcode.zdr`. The value is resolved per request, so toggling it
+`providerOptions.commandcode.zdr`. The value is resolved per request, so toggling it
 does not require restarting opencode.
 
 Precedence, highest first:
@@ -174,7 +190,6 @@ Precedence, highest first:
 1. `providerOptions.commandcode.zdr` (`true`/`false`), e.g. a model variant or agent option.
 2. An explicit `x-cmd-zdr` header in `opencode.json` `options.headers` or the call's `headers`.
 3. The toggle file (`zdr` key — see [Runtime toggles](#runtime-toggles)).
-4. `COMMANDCODE_ZDR=1|true|yes` in the environment.
 
 With tracing enabled, the trace logs a `zdr on|off` line per request so the toggle state is
 observable.
@@ -184,18 +199,32 @@ observable.
 `/cc-zdr` and `/cc-debug` flip the `zdr` and `debug` values in a small JSON file the provider
 reads on every request, so both settings change without restarting opencode. `/cc-status` shows
 the current state. The file is `~/.config/opencode/commandcode-toggles.json` (overridable with
-`COMMANDCODE_TOGGLES_FILE`); the environment variables above are used when a key is absent.
+`COMMANDCODE_TOGGLES_FILE` — see [Environment variables](#environment-variables)).
 
 These commands come from a small TUI plugin shipped in this package (`src/tui.tsx`). It is loaded
 from **source** — opencode compiles the TSX and provides the Solid runtime itself — so there is no
-build step and no runtime dependency for it. Register it once:
+build step and no runtime dependency for it. Register it once by adding the plugin file's absolute
+`file://` URL to the `plugin` list in `tui.json` (global `~/.config/opencode/tui.json`, or
+project-local `.opencode/tui.json`):
 
-```powershell
+```jsonc
-opencode plugin file:///C:/DevTools/pienv/ccprovider/src/tui.tsx
+// tui.json
+{
+  "plugin": ["file:///C:/dev/opencode-commandcode-provider/src/tui.tsx"]
+}
 ```
 
-That writes a `tui.json` `plugin` entry (project-local `.opencode/tui.json`, or global with
+Use an absolute path with forward slashes (`file:///C:/...` on Windows, `file:///home/...` on
-`--global`). Restart opencode after installing; then type `/cc-` for autocomplete. The plugin also
+Linux/macOS). Two rules for the entry:
+
+- It must point at **`src/tui.tsx` itself**. `opencode plugin <module>` cannot install this form:
+  the CLI requires a `package.json` manifest next to the target, so a bare `.tsx` must be wired in
+  by hand as above.
+- Never point `plugin` at the **repo root**. A directory spec makes opencode also treat the package
+  as a server plugin (via `package.json` `main` → `dist/index.js`), which is the AI SDK provider,
+  not a plugin — it loads as dead weight into every session.
+
+Restart opencode after registering the plugin; then type `/cc-` for autocomplete. The plugin also
 renders a **CommandCode panel in the session sidebar** (`zdr` / `debug`, plus live quota, above the
 built-in panels) that updates live as you toggle. Note that in opencode 1.x the slash menu and the
 `Ctrl+P` palette read the same command registry, so these entries appear in both. If you run opencode
@@ -224,7 +253,9 @@ instead of being shown as zero.
 The sidebar refreshes after each completed turn (`session.idle`, plus `session.status`,
 `session.updated`, and `message.updated` as fallbacks since `session.idle` is a server-plugin
 event that may never reach the TUI bus — all debounced) and when the active session changes,
-plus every 3 minutes as a fallback (`COMMANDCODE_QUOTA_INTERVAL_MS` overrides it) and the
+but never more than once every 2 minutes; every 5 minutes as a fallback
+(`COMMANDCODE_QUOTA_INTERVAL_MS` overrides the fallback interval — see
+[Environment variables](#environment-variables)) and the
 countdown ticks every 30 seconds. Key resolution and fetch failures are redacted and shown
 inline; the panel never stays on `loading…` and never blocks the provider. Extra trace lines
 (`tui-quota`) are appended to the debug log only when debug tracing is on.
@@ -257,7 +288,7 @@ The API key is never logged; quota errors pass through `redact()` like every oth
 | Finish reasons | Yes — unified (`stop`, `length`, `tool-calls`, `content-filter`, `error`, `other`) plus raw |
 | Sampling parameters | Yes — `temperature`, `topP`, `topK`, `stopSequences`, `seed`, presence/frequency penalties |
 | `reasoning_effort` | Yes — via `providerOptions.commandcode` |
-| Retry with backoff | Yes — 429/5xx and network errors, honouring `Retry-After` |
+| Retry with backoff | Yes — 429/5xx and network errors, honouring `Retry-After`; also retryable `error` events that arrive inside an HTTP 200 stream before any content (gateway 520s) |
 | Credential redaction | Yes — error bodies are scrubbed before surfacing |
 | Runtime toggles | Yes — `/cc-zdr` and `/cc-debug` flip the shared toggle file without a restart |
 | Sidebar status panel | Yes — a `sidebar_content` panel shows `zdr`/`debug` and live quota |
@@ -294,7 +325,7 @@ src/transform.ts  LanguageModelV3CallOptions -> /alpha/generate envelope (JSON s
 src/events.ts     NDJSON/SSE line iterator over the upstream response body.
 src/usage.ts      finish event -> V3 usage; finish-reason unification.
 src/redact.ts     Credential scrubbing for error surfaces.
-src/log.ts        Opt-in tracing (COMMANDCODE_DEBUG) to a log file.
+src/log.ts        Opt-in tracing (toggle file) to a log file.
 src/toggles.ts    Shared toggle file read by the provider and written by the TUI plugin.
 src/quota.ts      Live 5-hour/weekly/monthly quota from the alpha billing endpoints.
 src/tui.tsx       opencode TUI plugin: /cc-zdr, /cc-debug, /cc-status, /cc-usage + sidebar panel.
@@ -360,9 +391,10 @@ opencode run "Use the glob tool to list *.mjs and report the filenames." -m comm
 | `stream ended ... no finish` / truncated output | Upstream closed the connection early. The provider emits a synthetic `finish`, but the response is incomplete; retry the turn. |
 | Images are ignored by the model | The selected model is not vision-capable. Mark it with `"attachment": true` and `modalities.input: ["text","image"]` in `models`, and pick a vision model id. |
 | Upstream 400 about tool calls | A `tool-call` or `tool-result` without a matching pair slipped through. Pairing is enforced in `src/transform.ts`; report a repro if it still occurs. |
-| Need to see what the provider sends/receives | Set `COMMANDCODE_DEBUG=1` and read the appended log file (see [Debug tracing](#debug-tracing)). |
+| Need to see what the provider sends/receives | Flip `/cc-debug` and read the appended log file (see [Debug tracing](#debug-tracing)). |
 | `tool_choice` seemingly ignored | Expected for `required`; upstream cannot force a call. `none` and named-tool are emulated via the tool list. |
 | Config change had no effect | opencode reads config once at startup. Restart it. |
+| `/cc-*` commands and the sidebar panel are missing | TUI plugin not registered: add a direct `file://` URL to `src/tui.tsx` in `tui.json` (see [Runtime toggles](#runtime-toggles)). A repo-root or `src/tui.tsx`-less entry does not load. |
 
 ## Security
 

package.json

@@ -9,9 +9,6 @@
     ".": {
       "import": "./dist/index.js",
       "types": "./dist/index.d.ts"
-    },
-    "./tui": {
-      "import": "./src/tui.tsx"
     }
   },
   "files": [

scripts/sync-models.mjs

@@ -7,15 +7,16 @@
 // the result into the global opencode config. Deselecting a model removes it from the config.
 //
 // Run: node scripts/sync-models.mjs   (build first: npm run build)
-import { readFileSync, writeFileSync, renameSync } from "node:fs";
+import { mkdirSync, readFileSync, writeFileSync, renameSync } from "node:fs";
 import { homedir } from "node:os";
-import { join } from "node:path";
+import { dirname, join } from "node:path";
 import readline from "node:readline";
 import { createCommandCode } from "../dist/index.js";
 import { redact } from "../dist/redact.js";
 import { DEFAULT_BASE_URL, DEFAULT_CC_VERSION, DEFAULT_MAX_TOKENS, MODELS_PATH } from "../dist/constants.js";
 
 const CONFIG_PATH = join(homedir(), ".config", "opencode", "opencode.json");
+const CONSOLE_URL = "https://commandcode.ai/studio/provider";
 const CONCURRENCY = 6;
 
 // 64x64 solid red PNG, used only to provoke a vision-accepting vs vision-rejecting response.
@@ -36,8 +37,15 @@ const NO_IMAGE =
 const VISION_HEURISTIC = /claude|gpt-5|gemini|grok|qwen.*vl|vision|(?:^|[^a-z])vl(?:[^a-z]|$)|omni|multimodal/i;
 
 function readConfig(path) {
+  try {
     const text = readFileSync(path, "utf8");
-  return { text, data: JSON.parse(text) };
+    return { text, data: JSON.parse(text), exists: true };
+  } catch (error) {
+    // A fresh install has no opencode.json yet; treat it as empty so the console
+    // guidance / env-key path can still run. Bad JSON or unreadable files still throw.
+    if (error.code === "ENOENT") return { text: "", data: {}, exists: false };
+    throw error;
+  }
 }
 
 function detectIndent(text) {
@@ -360,23 +368,36 @@ function pickModels(catalog, existingIds) {
 }
 
 async function main() {
-  if (!process.stdin.isTTY || !process.stdout.isTTY) {
-    console.error("Interactive picker needs a TTY; nothing was written.");
-    process.exit(1);
-  }
-
   let config;
   let text;
+  let exists;
   try {
-    ({ text, data: config } = readConfig(CONFIG_PATH));
+    ({ text, data: config, exists } = readConfig(CONFIG_PATH));
   } catch (error) {
     console.error(`Cannot read ${CONFIG_PATH}: ${error instanceof Error ? error.message : String(error)}`);
     process.exit(1);
   }
 
+  // Mandatory first step: without a CommandCode key there is nothing to fetch or
+  // probe, so point a fresh user at the Console before doing anything else.
   const key = resolveKey(config, process.env);
   if (!key) {
-    console.error("No API key: set COMMANDCODE_API_KEY or provider.commandcode.options.apiKey");
+    const where = exists
+      ? `No CommandCode API key in ${CONFIG_PATH}.`
+      : `No opencode.json found at ${CONFIG_PATH}.`;
+    console.error(
+      `${where}\n` +
+        `Generate an API key in the CommandCode Console: ${CONSOLE_URL}\n` +
+        (exists
+          ? `Add it as provider.commandcode.options.apiKey, or export COMMANDCODE_API_KEY.`
+          : `Create that file with provider.commandcode.options.apiKey, or export COMMANDCODE_API_KEY.`) +
+        `\nNothing was written.`,
+    );
+    process.exit(1);
+  }
+
+  if (!process.stdin.isTTY || !process.stdout.isTTY) {
+    console.error("Interactive picker needs a TTY; nothing was written.");
     process.exit(1);
   }
 
@@ -447,7 +468,9 @@ async function main() {
   config.provider.commandcode.models = merged;
 
   const indent = detectIndent(text);
-  const trailing = text.endsWith("\n") ? "\n" : "";
+  const trailing = text === "" || text.endsWith("\n") ? "\n" : "";
+  // The config directory may not exist yet on a fresh install; create it before the atomic write.
+  mkdirSync(dirname(CONFIG_PATH), { recursive: true });
   writeAtomic(CONFIG_PATH, `${JSON.stringify(config, null, indent)}${trailing}`);
 
   console.log(

src/log.ts

@@ -1,4 +1,4 @@
-// Opt-in tracing to a log file. Gated on COMMANDCODE_DEBUG=1 or the shared toggle file;
+// Opt-in tracing to a log file. Gated on the shared toggle file (debug key);
 // silent (and no file created) by default. Uses only Node built-ins so the built output
 // stays dependency-free.
 
@@ -6,9 +6,9 @@ import { appendFileSync } from "node:fs";
 import { tmpdir } from "node:os";
 import { join } from "node:path";
 import { redact } from "./redact.js";
-import { debugFromEnvOrFile } from "./toggles.js";
+import { debugEnabled } from "./toggles.js";
 
-const FILE = process.env["COMMANDCODE_DEBUG_FILE"] ?? join(tmpdir(), "commandcode-debug.log");
+const FILE = join(tmpdir(), "commandcode-debug.log");
 
 function serialize(value: unknown): string {
   if (typeof value === "string") return value;
@@ -34,14 +34,14 @@ function emit(scope: string, args: unknown[]): void {
   }
 }
 
-/** Effective default tracing state (env var or toggle file). Re-read on every call. */
+/** Effective default tracing state (toggle file). Re-read on every call. */
 export function isDebugEnabled(): boolean {
-  return debugFromEnvOrFile();
+  return debugEnabled();
 }
 
 /** Write a trace line (appended) if tracing is enabled. Values are redacted before write. */
 export function debug(scope: string, ...args: unknown[]): void {
-  if (!debugFromEnvOrFile()) return;
+  if (!debugEnabled()) return;
   emit(scope, args);
 }
 

src/model.ts

@@ -21,12 +21,12 @@ import {
 import { iterateEvents } from "./events.js";
 import { debugWhen, isDebugEnabled } from "./log.js";
 import { redact } from "./redact.js";
-import { zdrFromEnvOrFile } from "./toggles.js";
+import { zdrEnabled } from "./toggles.js";
 import { transform } from "./transform.js";
 import { finishReasonFrom, usageFromFinish, costFromFinishStep, costFromProviderMetadata, type FinishEvent, type FinishStepEvent, type ProviderMetadataEvent } from "./usage.js";
 
 // x-cmd-zdr breaks some models; off by default (server.py never sends it).
-// Precedence: providerOptions.commandcode.zdr > x-cmd-zdr header > toggle file > COMMANDCODE_ZDR.
+// Precedence: providerOptions.commandcode.zdr > x-cmd-zdr header > toggle file.
 // "undefined" means the header is omitted entirely.
 function providerScoped(options: LanguageModelV3CallOptions): Record<string, unknown> | undefined {
   return options.providerOptions?.["commandcode"] as Record<string, unknown> | undefined;
@@ -44,7 +44,7 @@ function resolveZdr(
   if (typeof scoped === "boolean") return scoped ? "1" : "0";
   const header = headerValue({ ...providerHeaders, ...(options.headers ?? {}) }, "x-cmd-zdr");
   if (header !== undefined) return truthyHeader(header) ? "1" : "0";
-  return zdrFromEnvOrFile() ? "1" : undefined;
+  return zdrEnabled() ? "1" : undefined;
 }
 
 function resolveDebug(options: LanguageModelV3CallOptions): boolean {
@@ -109,6 +109,50 @@ function isRetryable(status: number): boolean {
   return status === 429 || (status >= 500 && status < 600);
 }
 
+// Upstream can answer HTTP 200 and then fail inside the stream (SSE `error` event).
+// CommandCode's gateway marks these with `statusCode`/`isRetryable`; retry before content.
+type StreamErrorInfo = { message: string; statusCode?: number; isRetryable?: boolean };
+
+function streamErrorInfo(evt: Record<string, any>): StreamErrorInfo | null {
+  if (evt?.["type"] !== "error") return null;
+  const err = evt["error"];
+  if (err === null || typeof err !== "object") return { message: String(err) };
+  const record = err as Record<string, unknown>;
+  const message = typeof record["message"] === "string" ? record["message"] : JSON.stringify(err);
+  return {
+    message,
+    ...(typeof record["statusCode"] === "number" ? { statusCode: record["statusCode"] } : {}),
+    ...(typeof record["isRetryable"] === "boolean" ? { isRetryable: record["isRetryable"] } : {}),
+  };
+}
+
+function isRetryableStreamError(info: StreamErrorInfo): boolean {
+  return info.isRetryable === true || (info.statusCode !== undefined && isRetryable(info.statusCode));
+}
+
+// Only `start`/`start-step` are safe to buffer while looking for an immediate stream error;
+// once real output appears a retry would duplicate the response.
+const STREAM_PREAMBLE = new Set(["start", "start-step"]);
+const MAX_PEEK_EVENTS = 32;
+
+type OpenedStream = {
+  response: Response;
+  prefix: Record<string, any>[];
+  events: AsyncGenerator<Record<string, any>>;
+};
+
+async function* emptyEvents(): AsyncGenerator<Record<string, any>> {
+  /* no-op */
+}
+
+async function* replayEvents(
+  prefix: Record<string, any>[],
+  rest: AsyncGenerator<Record<string, any>>,
+): AsyncGenerator<Record<string, any>> {
+  for (const evt of prefix) yield evt;
+  for await (const evt of rest) yield evt;
+}
+
 function retryDelay(attempt: number, retryAfter: string | null, maxDelay: number): number {
   if (retryAfter) {
     const seconds = Number(retryAfter);
@@ -186,7 +230,7 @@ class CommandCodeLanguageModel implements LanguageModelV3 {
     body: string,
     options: LanguageModelV3CallOptions,
     dbg: boolean,
-  ): Promise<Response> {
+  ): Promise<OpenedStream> {
     const headers = this.requestHeaders(options.headers, resolveZdr(options, this.opts.headers), dbg);
     let lastError: unknown;
     const url = `${this.opts.baseURL}${GENERATE_PATH}`;
@@ -247,8 +291,60 @@ class CommandCodeLanguageModel implements LanguageModelV3 {
           continue;
         }
       }
+      if (!response.ok || !response.body) {
         debugWhen(dbg, "fetch", "response", `url=${url}`, `attempt=${attempt}`, `status=${status}`, `elapsedMs=${Date.now() - started}`);
-      return response;
+        return { response, prefix: [], events: emptyEvents() };
+      }
+
+      // HTTP 200 can still hide an upstream failure. Buffer only the preamble; if the first
+      // real event is a retryable stream error, retry before anything reaches the client.
+      const events = iterateEvents(response.body, dbg);
+      const prefix: Record<string, any>[] = [];
+      let streamRetry: StreamErrorInfo | undefined;
+      while (prefix.length < MAX_PEEK_EVENTS) {
+        const next = await events.next();
+        if (next.done) break;
+        const evt = next.value;
+        prefix.push(evt);
+        const info = streamErrorInfo(evt);
+        if (info) {
+          if (isRetryableStreamError(info) && attempt < this.opts.maxRetries) streamRetry = info;
+          break;
+        }
+        if (!STREAM_PREAMBLE.has(evt["type"])) break;
+      }
+
+      if (streamRetry) {
+        const wait = retryDelay(attempt, null, this.opts.retryMaxDelay);
+        debugWhen(
+          dbg,
+          "fetch",
+          "stream-retry",
+          `url=${url}`,
+          `attempt=${attempt}`,
+          `status=${streamRetry.statusCode ?? "unknown"}`,
+          `message=${redact(streamRetry.message)}`,
+          `waitMs=${wait}`,
+        );
+        try {
+          await events.return(undefined as never);
+        } catch {
+          /* ignore */
+        }
+        try {
+          await response.body.cancel();
+        } catch {
+          /* ignore */
+        }
+        if (options.abortSignal?.aborted) {
+          throw options.abortSignal.reason ?? new Error("Aborted");
+        }
+        await sleep(wait);
+        continue;
+      }
+
+      debugWhen(dbg, "fetch", "response", `url=${url}`, `attempt=${attempt}`, `status=${status}`, `elapsedMs=${Date.now() - started}`);
+      return { response, prefix, events };
     }
 
     throw lastError instanceof Error ? lastError : new Error("Upstream unreachable");
@@ -279,11 +375,11 @@ class CommandCodeLanguageModel implements LanguageModelV3 {
     const dbg = resolveDebug(options);
     const body = transform(options, this.modelId);
     debugWhen(dbg, "doStream", `model=${this.modelId}`, `bodyBytes=${Buffer.byteLength(body, "utf8")}`);
-    const response = await this.fetchWithRetry(body, options, dbg);
+    const { response, prefix, events } = await this.fetchWithRetry(body, options, dbg);
     if (!response.ok || !response.body) throw await this.errorFrom(response, dbg);
     debugWhen(dbg, "doStream", "ok", `status=${response.status}`);
 
-    const stream = toReadableStream(this.streamParts(response.body, dbg));
+    const stream = toReadableStream(this.streamParts(prefix, events, dbg));
     return {
       stream,
       request: { body: JSON.parse(body) as unknown },
@@ -357,7 +453,11 @@ class CommandCodeLanguageModel implements LanguageModelV3 {
     return { content, finishReason, usage, warnings, request, response, ...(providerMetadata ? { providerMetadata } : {}) };
   }
 
-  private async *streamParts(body: ReadableStream<Uint8Array>, dbg: boolean): AsyncGenerator<LanguageModelV3StreamPart> {
+  private async *streamParts(
+    prefix: Record<string, any>[],
+    events: AsyncGenerator<Record<string, any>>,
+    dbg: boolean,
+  ): AsyncGenerator<LanguageModelV3StreamPart> {
     yield { type: "stream-start", warnings: [] };
 
     const textId = "text-0";
@@ -372,7 +472,7 @@ class CommandCodeLanguageModel implements LanguageModelV3 {
     let marketCost: number | undefined;
     let errored = false;
 
-    for await (const evt of iterateEvents(body, dbg)) {
+    for await (const evt of replayEvents(prefix, events)) {
       const payload = JSON.stringify(evt);
       debugWhen(dbg, "stream", `event=${evt.type}`, `payload=${payload && payload.length > 4096 ? payload.slice(0, 4096) + "…" : payload}`);
       switch (evt.type) {

src/toggles.ts

@@ -45,17 +45,11 @@ export function toggle(name: ToggleName): Toggles {
   return writeToggles({ [name]: current[name] !== true });
 }
 
-export function envFlag(name: string): boolean {
+/** Toggle state from the shared flag file; an absent key means off. */
-  return /^(1|true|yes)$/i.test(process.env[name] ?? "");
+export function zdrEnabled(): boolean {
+  return readToggles().zdr === true;
 }
 
-/** Flag file wins over the environment; unset file falls back to the env var. */
+export function debugEnabled(): boolean {
-export function zdrFromEnvOrFile(): boolean {
+  return readToggles().debug === true;
-  const file = readToggles().zdr;
-  return typeof file === "boolean" ? file : envFlag("COMMANDCODE_ZDR");
-}
-
-export function debugFromEnvOrFile(): boolean {
-  const file = readToggles().debug;
-  return typeof file === "boolean" ? file : envFlag("COMMANDCODE_DEBUG");
 }

src/tui.tsx

@@ -21,7 +21,7 @@ import { createSignal } from "solid-js";
 
 import { fetchQuota, formatQuota, formatReset, percent, quotaBar, type QuotaResult, type QuotaWindow } from "./quota.js";
 import { redact } from "./redact.js";
-import { debugFromEnvOrFile, readToggles, toggle, type ToggleName, type Toggles } from "./toggles.js";
+import { debugEnabled, readToggles, toggle, type ToggleName, type Toggles } from "./toggles.js";
 
 type ToastVariant = "info" | "success" | "warning" | "error";
 
@@ -69,9 +69,11 @@ type TuiApi = {
 const ID = "commandcode-toggles";
 const CATEGORY = "CommandCode";
 const SIDEBAR_ORDER = 90;
-const DEFAULT_QUOTA_INTERVAL_MS = 180_000;
+const DEFAULT_QUOTA_INTERVAL_MS = 300_000;
+const MIN_REFRESH_INTERVAL_MS = 120_000;
 // Extra TUI-bus signals that mark the end of a turn. `session.idle` is a server
 // plugin event and may never fire here; these keep the panel fresh regardless.
+// All event-driven refreshes are floored to MIN_REFRESH_INTERVAL_MS (hard skip).
 const QUOTA_TRIGGER_EVENTS = ["session.idle", "session.status", "session.updated", "message.updated"] as const;
 
 function safeEnv(name: string): string | undefined {
@@ -84,8 +86,8 @@ function safeEnv(name: string): string | undefined {
 
 /** Debug-only trace to the shared debug file. Silent unless debug is on; never logs the key. */
 function trace(...args: unknown[]): void {
-  if (!debugFromEnvOrFile()) return;
+  if (!debugEnabled()) return;
-  const file = safeEnv("COMMANDCODE_DEBUG_FILE") ?? join(tmpdir(), "commandcode-debug.log");
+  const file = join(tmpdir(), "commandcode-debug.log");
   const parts = args.map((value) => (typeof value === "string" ? value : safeJson(value)));
   const line = `[commandcode] ${new Date().toISOString()} [tui-quota] ${parts.join(" ")}`;
   try {
@@ -168,6 +170,11 @@ function quotaIntervalMs(): number {
   return Number.isFinite(raw) && raw > 0 ? raw : DEFAULT_QUOTA_INTERVAL_MS;
 }
 
+function formatClock(ms: number): string {
+  if (!(ms > 0)) return "--:--:--";
+  return new Date(ms).toTimeString().slice(0, 8);
+}
+
 function shortWindow(w: QuotaWindow, nowMs: number): string {
   const label = w.id === "fiveHour" ? "5h" : w.id === "weekly" ? "7d" : "mo";
   const reset = formatReset(w.resetAtMs, nowMs);
@@ -180,9 +187,11 @@ export const tui = async (api: TuiApi): Promise<void> => {
   const [toggles, setToggles] = createSignal(readToggles());
   const [quota, setQuota] = createSignal<QuotaResult | null>(null);
   const [now, setNow] = createSignal(Date.now());
+  const [updatedAt, setUpdatedAt] = createSignal(0);
 
   let inflight = false;
   let pending = false;
+  let lastRefreshAt = 0;
   let debounceTimer: ReturnType<typeof setTimeout> | undefined;
   let lastSessionId: string | undefined;
 
@@ -194,6 +203,11 @@ export const tui = async (api: TuiApi): Promise<void> => {
       trace("coalesced", "inflight");
       return;
     }
+    const since = Date.now() - lastRefreshAt;
+    if (lastRefreshAt > 0 && since < MIN_REFRESH_INTERVAL_MS) {
+      trace("rate-limited", `remainingMs=${MIN_REFRESH_INTERVAL_MS - since}`);
+      return;
+    }
     inflight = true;
     try {
       const { key: apiKey, source } = resolveApiKeySource(api);
@@ -222,7 +236,9 @@ export const tui = async (api: TuiApi): Promise<void> => {
       trace("fetch-throw", message);
     } finally {
       inflight = false;
+      lastRefreshAt = Date.now();
       setNow(Date.now());
+      setUpdatedAt(Date.now());
       if (pending) {
         pending = false;
         trace("flush-pending");
@@ -284,6 +300,7 @@ export const tui = async (api: TuiApi): Promise<void> => {
         ) : quota() === null ? (
           <text fg={theme.current.textMuted}>quota: loading…</text>
         ) : null}
+        <text fg={theme.current.textMuted}>updated @ {formatClock(updatedAt())}</text>
       </box>
     );
   };
@@ -365,6 +382,7 @@ export const tui = async (api: TuiApi): Promise<void> => {
           }
           setQuota(result);
           setNow(Date.now());
+          setUpdatedAt(Date.now());
           api.ui.toast({ title: CATEGORY, message: formatQuota(result.quota), variant: "info", duration: 15000 });
         },
       },