@hank-warren/pi-auto-permissions 0.7.2 → 0.9.0

package/CHANGELOG.md

@@ -1,5 +1,40 @@
 # @hank-warren/pi-auto-permissions
 
+## 0.9.0
+
+### Minor Changes
+
+- 7d96995: Stop `pi-herdr-auto-title` depending on `pi-auto-permissions`.
+
+  The pane-title extension deep-imported `guardian-transport.ts` — roughly forty lines resolving `completeSimple` through the host `ModelRuntime` so provider extensions (Anthropic OAuth shaping) apply to background model calls. Reusing it cost an entire dependency on the permissions engine: 125 kB unpacked in `node_modules`, a transitive `pi-permission-selector`, and a no-op release of the title extension every time the guardian shipped.
+
+  The file is now duplicated byte-for-byte into both packages, and `scripts/validate.py` fails if the copies drift. `pi-herdr-auto-title` declares no dependencies at all and its tarball is unchanged apart from the added source file.
+
+  To keep the copies identical, `resolveGuardianCompleteSimple` takes the calling package's name as a second argument and uses it in the compat-fallback warning, replacing the hardcoded `[pi-auto-permissions]` prefix. That is a breaking change only for an external deep import of `@hank-warren/pi-auto-permissions/guardian-transport.ts`; the extension's own behavior is unchanged.
+
+### Patch Changes
+
+- 7d96995: Make every declared dependency accurate.
+
+  `pi-permission-selector` no longer declares a `@earendil-works/pi-coding-agent` peer dependency. It imports nothing from it — the claim was left over from 1.0.0, which removed the extension and the selector monkey patch and turned the package into a pure library.
+
+  `pi-auto-permissions` floors `@hank-warren/pi-permission-selector` at `^1.1.0` and `pi-plan-mode` floors `@narumitw/pi-tui-kit` at `^0.49.3`, in both cases the lowest version the package is actually tested against. Neither uses a feature the old floor lacked, so nothing changes at install time under caret resolution; the ranges simply no longer claim support that nothing verifies.
+
+  `scripts/validate.py` now enforces that a package's Pi `peerDependencies` are exactly the Pi packages its shipped sources import, in both directions.
+
+- Updated dependencies [7d96995]
+  - @hank-warren/pi-permission-selector@1.1.1
+
+## 0.8.0
+
+### Minor Changes
+
+- 52ff06e: Add a dedicated `OpenAI Codex - Auto Permissions` provider with its own OAuth credential slot, allowing guardian reviews to use a separate ChatGPT account without replacing the main OpenAI Codex login.
+
+### Patch Changes
+
+- 312da9a: Add a pulsing warning light to interactive approval prompts so they remain visually distinct from the surrounding transcript.
+
 ## 0.7.2
 
 ### Patch Changes

package/README.md

@@ -114,12 +114,18 @@ Set `enabled` to `false` to disable the extension. Invalid configuration fails c
 
 ## Guardian configuration
 
-By default, the guardian uses Pi's active model with low reasoning effort and a 30-second timeout. You can select a separate low-cost model:
+By default, the guardian uses Pi's active model with low reasoning effort and a 30-second timeout. You can select a separate low-cost model.
+
+For OpenAI Codex, this package registers a second provider named **OpenAI Codex - Auto Permissions**. Pi keys OAuth credentials by provider id, so this provider can use a different ChatGPT account without replacing the normal `openai-codex` login used by your main agent.
+
+1. Run `/login` and select **OpenAI Codex - Auto Permissions**.
+2. Complete OAuth with the account reserved for guardian reviews.
+3. Select the dedicated provider in the Auto Permissions config:
 
 ```json
 {
   "reviewer": {
-    "provider": "openai-codex",
+    "provider": "openai-codex-auto-permissions",
     "model": "gpt-5.6-luna",
     "reasoningEffort": "low",
     "timeoutMs": 30000
@@ -127,6 +133,10 @@ By default, the guardian uses Pi's active model with low reasoning effort and a
 }
 ```
 
+The separate credential is stored in Pi's normal `auth.json` under `openai-codex-auto-permissions`; the existing `openai-codex` credential is unchanged. The alias uses Pi's built-in OpenAI Codex OAuth flow, model catalog, and transport. Because it is a normal provider, its models also appear under the dedicated provider name in Pi's model selector after login.
+
+Other reviewer providers continue to work by setting their normal provider and model ids.
+
 ### Guardian prompt
 
 The bundled prompt evaluates authorization, command risk, and compliance with user constraints. Replace it inline with `systemPrompt`, or load a file:
@@ -238,10 +248,10 @@ The path defaults to `usage.jsonl` beside the Auto Permissions config and resolv
 The default UI shows guardian progress as a single animated status line in a temporary widget above the editor:
 
 ```text
-auto permissions · Git commit · ✶ waiting for openai-codex/gpt-5.6-luna
+auto permissions · Git commit · ✶ waiting for openai-codex-auto-permissions/gpt-5.6-luna
 ```
 
-A sparkle spinner (`✶ ✸ ✻ ✽`) cycles while the guardian is reviewing and resolves to `✓ approved`, `↻ revision requested`, `? waiting for your approval`, or `✗ blocked`. The guardian's reason, when present, appears on a dim second line; the command itself is not repeated because it is already visible in the Bash tool box. Configure the widget with:
+A sparkle spinner (`✶ ✸ ✻ ✽`) cycles while the guardian is reviewing and resolves to `✓ approved`, `↻ revision requested`, `? waiting for your approval`, or `✗ blocked`. When approval is needed, the selector's leading `●` pulses between warning-bright and dim so the active prompt remains visually distinct from the transcript. The guardian's reason, when present, appears on a dim second line; the command itself is not repeated because it is already visible in the Bash tool box. Configure the widget with:
 
 ```json
 {
@@ -257,7 +267,7 @@ Set `placement` to `toolRow` to show the review inside Pi's Bash tool row:
 
 ```text
 $ git commit --dry-run -m "fix auth"
-  ◌ guardian running · Git commit · openai-codex/gpt-5.6-luna
+  ◌ guardian running · Git commit · openai-codex-auto-permissions/gpt-5.6-luna
 ```
 
 `toolRow` reconstructs Pi's standard local Bash definition because Pi does not expose renderer-only decoration. Do not use it with SDK-provided, remote, sandboxed, or otherwise replaced Bash backends. The extension detects non-native Bash tools and falls back to the widget instead of replacing them.

package/approval-pulse.ts

@@ -0,0 +1,80 @@
+import { truncateToWidth, type Component } from "@earendil-works/pi-tui";
+
+export const APPROVAL_PULSE_INTERVAL_MS = 600;
+
+export interface ApprovalPulseTheme {
+  fg(role: "warning" | "dim", text: string): string;
+  bold(text: string): string;
+}
+
+export interface ApprovalPulseScheduler {
+  setInterval(callback: () => void, intervalMs: number): unknown;
+  clearInterval(handle: unknown): void;
+}
+
+type InteractiveComponent = Component & Required<Pick<Component, "handleInput">>;
+
+const DEFAULT_SCHEDULER: ApprovalPulseScheduler = {
+  setInterval(callback, intervalMs) {
+    const timer = setInterval(callback, intervalMs);
+    timer.unref?.();
+    return timer;
+  },
+  clearInterval(handle) {
+    clearInterval(handle as ReturnType<typeof setInterval>);
+  },
+};
+
+/**
+ * Decorates an interactive prompt with a bright/dim approval light.
+ *
+ * The child still owns all content and input. Only its first rendered line is
+ * replaced, preserving the selector's existing spacing and wrapping exactly.
+ */
+export class PulsingApprovalIndicator implements Component {
+  private bright = true;
+  private timer: unknown;
+
+  constructor(
+    private readonly headline: string,
+    private readonly child: InteractiveComponent,
+    private readonly theme: ApprovalPulseTheme,
+    private readonly requestRender: () => void,
+    private readonly scheduler: ApprovalPulseScheduler = DEFAULT_SCHEDULER,
+  ) {}
+
+  start(): void {
+    if (this.timer !== undefined) return;
+    this.timer = this.scheduler.setInterval(() => {
+      this.bright = !this.bright;
+      this.requestRender();
+    }, APPROVAL_PULSE_INTERVAL_MS);
+  }
+
+  stop(): void {
+    if (this.timer === undefined) return;
+    this.scheduler.clearInterval(this.timer);
+    this.timer = undefined;
+  }
+
+  isRunning(): boolean {
+    return this.timer !== undefined;
+  }
+
+  render(width: number): string[] {
+    const lines = this.child.render(width);
+    const glyph = this.bright
+      ? this.theme.fg("warning", this.theme.bold("●"))
+      : this.theme.fg("dim", "●");
+    const headline = truncateToWidth(`${glyph} ${this.headline}`, Math.max(1, width), "");
+    return lines.length > 0 ? [headline, ...lines.slice(1)] : [headline];
+  }
+
+  handleInput(data: string): void {
+    this.child.handleInput(data);
+  }
+
+  invalidate(): void {
+    this.child.invalidate();
+  }
+}

package/guardian-transport.ts

@@ -1,17 +1,17 @@
 import { completeSimple } from "@earendil-works/pi-ai/compat";
 
 /**
- * Guardian model dispatch seam.
+ * Model dispatch seam for extension-issued background model calls.
  *
  * pi-ai's `compat.completeSimple` dispatches through pi-ai's own api registry,
  * which extension-registered provider transports never reach: pi routes
  * `pi.registerProvider()` configs only through the coding agent's
  * `ModelRuntime` -> `provider-composer.streamWith`. Auth-shaping extensions
  * such as `@gotgenes/pi-anthropic-auth` install their transport wrapper there,
- * so a reviewer request issued via `compat.completeSimple` reaches Anthropic
- * unshaped, and Claude Pro/Max OAuth requests fail with the misleading
- * "extra usage" HTTP 400 (see gotgenes/pi-anthropic-auth
- * docs/architecture.md, "The remaining gap: pi-ai compat dispatch").
+ * so a request issued via `compat.completeSimple` reaches Anthropic unshaped,
+ * and Claude Pro/Max OAuth requests fail with the misleading "extra usage"
+ * HTTP 400 (see gotgenes/pi-anthropic-auth docs/architecture.md, "The
+ * remaining gap: pi-ai compat dispatch").
  *
  * `ExtensionContext.modelRegistry` is a deliberately narrowed facade over
  * `ModelRuntime` that re-exports `complete()` but not `completeSimple()` —
@@ -21,12 +21,24 @@ import { completeSimple } from "@earendil-works/pi-ai/compat";
  * marked `private` in the .d.ts), and fall back to compat dispatch when a
  * future pi release renames the seam — degrading to the historical behavior
  * instead of crashing.
+ *
+ * THIS FILE IS DUPLICATED VERBATIM across packages/pi-auto-permissions and
+ * packages/pi-herdr-auto-title, and `scripts/validate.py` fails if the copies
+ * differ by a single byte. The duplication is deliberate: sharing it made a
+ * pane-titling extension depend on an entire permissions engine to reuse a
+ * dozen lines. Edit one copy, then copy it over the other. Every
+ * package-specific detail must stay behind a parameter — hence `packageLabel`.
+ * The "guardian" in the names is historical: pi-auto-permissions' guardian
+ * reviewer was the first caller.
  */
 export type GuardianCompleteSimple = typeof completeSimple;
 
 let warnedCompatFallback = false;
 
-export function resolveGuardianCompleteSimple(modelRegistry: unknown): GuardianCompleteSimple {
+export function resolveGuardianCompleteSimple(
+  modelRegistry: unknown,
+  packageLabel: string,
+): GuardianCompleteSimple {
   const runtime = (modelRegistry as { runtime?: unknown } | null | undefined)?.runtime;
   const candidate = (runtime as { completeSimple?: unknown } | null | undefined)?.completeSimple;
   if (typeof candidate === "function") {
@@ -35,7 +47,7 @@ export function resolveGuardianCompleteSimple(modelRegistry: unknown): GuardianC
   if (!warnedCompatFallback) {
     warnedCompatFallback = true;
     console.warn(
-      "[pi-auto-permissions] host ModelRuntime completeSimple is unavailable; falling back to pi-ai compat dispatch. Extension-registered provider transports (e.g. Anthropic OAuth shaping) will not apply to reviewer calls.",
+      `[${packageLabel}] host ModelRuntime completeSimple is unavailable; falling back to pi-ai compat dispatch. Extension-registered provider transports (e.g. Anthropic OAuth shaping) will not apply to this extension's model calls.`,
     );
   }
   return completeSimple;

package/index.ts

@@ -1,5 +1,9 @@
 import { cleanupSessionResources, type Message } from "@earendil-works/pi-ai";
 import { resolveGuardianCompleteSimple } from "./guardian-transport.js";
+import {
+  createAutoPermissionsOpenAICodexProvider,
+  isOpenAICodexModel,
+} from "./openai-codex-provider.js";
 import { detectSubagentContext } from "./subagent-context.js";
 import * as PiCodingAgent from "@earendil-works/pi-coding-agent";
 import type { ExtensionAPI, ExtensionContext } from "@earendil-works/pi-coding-agent";
@@ -38,6 +42,7 @@ import {
 } from "./review.js";
 import { mergeOverrideEvidence, type PermissionOverride } from "./override-evidence.js";
 import { OptionSelector } from "@hank-warren/pi-permission-selector/selector.ts";
+import { PulsingApprovalIndicator } from "./approval-pulse.js";
 
 const WIDGET_KEY = "auto-permissions";
 const PROJECT_CONFIG_DIR_NAME = (PiCodingAgent as { CONFIG_DIR_NAME?: string }).CONFIG_DIR_NAME ?? ".pi";
@@ -125,8 +130,8 @@ function createUuidV7(): string {
   return `${hex.slice(0, 8)}-${hex.slice(8, 12)}-${hex.slice(12, 16)}-${hex.slice(16, 20)}-${hex.slice(20)}`;
 }
 
-function reviewerSessionId(provider: string): string {
-  return provider === "openai-codex" ? createUuidV7() : `ap-review-${randomUUID()}`;
+function reviewerSessionId(model: { api?: string }): string {
+  return isOpenAICodexModel(model) ? createUuidV7() : `ap-review-${randomUUID()}`;
 }
 
 function loadNativeBashOptions(cwd: string, projectTrusted: boolean): { commandPrefix?: string; shellPath?: string } {
@@ -359,7 +364,7 @@ export default function autoPermissionsExtension(pi: ExtensionAPI) {
       throw new Error("compact review evidence exceeds the review model's safe context budget");
     }
 
-    const sessionId = base?.sessionId ?? reviewerSessionId(model.provider);
+    const sessionId = base?.sessionId ?? reviewerSessionId(model);
     const attemptGeneration = reviewerGeneration;
     activeReviewerSessionId = sessionId;
     const timeoutSignal = AbortSignal.timeout(config.reviewer?.timeoutMs ?? 30_000);
@@ -378,7 +383,7 @@ export default function autoPermissionsExtension(pi: ExtensionAPI) {
       }
       // Dispatch through the host ModelRuntime so extension-registered provider
       // transports (e.g. pi-anthropic-auth OAuth shaping) apply; see guardian-transport.ts.
-      const response = await resolveGuardianCompleteSimple(ctx.modelRegistry)(
+      const response = await resolveGuardianCompleteSimple(ctx.modelRegistry, "pi-auto-permissions")(
         model,
         { systemPrompt, messages },
         {
@@ -388,7 +393,7 @@ export default function autoPermissionsExtension(pi: ExtensionAPI) {
           signal: reviewSignal,
           reasoning: config.reviewer?.reasoningEffort ?? "low",
           sessionId,
-          transport: model.provider === "openai-codex" ? "websocket" : "auto",
+          transport: isOpenAICodexModel(model) ? "websocket" : "auto",
           cacheRetention: "long",
         },
       );
@@ -677,9 +682,14 @@ export default function autoPermissionsExtension(pi: ExtensionAPI) {
     signal: AbortSignal,
     opts: { allowComment: boolean },
   ): Promise<string | undefined> {
+    let indicator: PulsingApprovalIndicator | undefined;
     return ctx.ui.custom<string | undefined>((tui, theme, _keybindings, done) => {
+      let finished = false;
       const onAbort = () => finish(undefined);
       const finish = (result: string | undefined) => {
+        if (finished) return;
+        finished = true;
+        indicator?.stop();
         signal.removeEventListener("abort", onAbort);
         done(result);
       };
@@ -701,12 +711,19 @@ export default function autoPermissionsExtension(pi: ExtensionAPI) {
         onCancel: () => finish(undefined),
         requestRender: () => tui.requestRender(),
       });
+      indicator = new PulsingApprovalIndicator(
+        title.split("\n", 1)[0] ?? "Auto Permissions needs approval",
+        selector,
+        theme,
+        () => tui.requestRender(),
+      );
+      signal.addEventListener("abort", onAbort, { once: true });
       // `done` before the factory resolves is safe: pi marks the dialog closed
       // and never mounts the component.
       if (signal.aborted) onAbort();
-      else signal.addEventListener("abort", onAbort, { once: true });
-      return selector;
-    });
+      else indicator.start();
+      return indicator;
+    }).finally(() => indicator?.stop());
   }
 
   async function askUser(
@@ -980,6 +997,13 @@ export default function autoPermissionsExtension(pi: ExtensionAPI) {
   });
 
   pi.on("session_start", async (_event, ctx) => {
+    const openaiCodex = ctx.modelRegistry.getProvider("openai-codex");
+    if (openaiCodex?.auth.oauth) {
+      pi.registerProvider(createAutoPermissionsOpenAICodexProvider(openaiCodex));
+    } else {
+      console.warn("[pi-auto-permissions] OpenAI Codex OAuth is unavailable; the dedicated Auto Permissions login was not registered.");
+    }
+
     reviewerLifecycleController.abort();
     discardReviewerLineage();
     reviewerLifecycleController = new AbortController();

package/openai-codex-provider.ts

@@ -0,0 +1,42 @@
+import type { Provider } from "@earendil-works/pi-ai";
+
+export const AUTO_PERMISSIONS_OPENAI_CODEX_PROVIDER_ID = "openai-codex-auto-permissions";
+export const AUTO_PERMISSIONS_OPENAI_CODEX_PROVIDER_NAME = "OpenAI Codex - Auto Permissions";
+
+/**
+ * Registering a second provider id gives Pi a second auth.json credential slot.
+ * The transport and model catalog still come from Pi's built-in OpenAI Codex
+ * provider, so the alias follows the host Pi version instead of duplicating its
+ * OAuth or Responses implementation here. The effective catalog is snapshotted
+ * when the session starts; the source provider remains the sole owner of any
+ * dynamic catalog refresh state.
+ */
+export function createAutoPermissionsOpenAICodexProvider(source: Provider): Provider {
+  const models = source.getModels().map((model) => ({
+    ...model,
+    provider: AUTO_PERMISSIONS_OPENAI_CODEX_PROVIDER_ID,
+  }));
+  const oauth = source.auth.oauth;
+
+  if (!oauth) {
+    throw new Error("Pi's OpenAI Codex provider does not expose OAuth authentication");
+  }
+
+  return {
+    ...source,
+    id: AUTO_PERMISSIONS_OPENAI_CODEX_PROVIDER_ID,
+    name: AUTO_PERMISSIONS_OPENAI_CODEX_PROVIDER_NAME,
+    auth: {
+      oauth: {
+        ...oauth,
+        name: AUTO_PERMISSIONS_OPENAI_CODEX_PROVIDER_NAME,
+      },
+    },
+    getModels: () => models,
+    refreshModels: undefined,
+  };
+}
+
+export function isOpenAICodexModel(model: { api?: string }): boolean {
+  return model.api === "openai-codex-responses";
+}

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@hank-warren/pi-auto-permissions",
-  "version": "0.7.2",
+  "version": "0.9.0",
   "description": "Fork of @ogulcancelik/pi-auto-permissions with dialog-answer user evidence: context-aware Bash permissions for Pi with automated guardian review.",
   "type": "module",
   "keywords": [
@@ -39,8 +39,10 @@
   "files": [
     "NOTICE.md",
     "index.ts",
+    "approval-pulse.ts",
     "config.ts",
     "guardian-transport.ts",
+    "openai-codex-provider.ts",
     "subagent-context.ts",
     "evaluation-log.ts",
     "gates.ts",
@@ -53,7 +55,7 @@
     "CHANGELOG.md"
   ],
   "dependencies": {
-    "@hank-warren/pi-permission-selector": "^1.0.1"
+    "@hank-warren/pi-permission-selector": "^1.1.1"
   },
   "peerDependencies": {
     "@earendil-works/pi-ai": "*",