pi-oracle 0.2.2 → 0.3.1

package/CHANGELOG.md

@@ -1,5 +1,26 @@
 # Changelog
 
+## 0.3.1 - 2026-04-08
+
+### Changed
+- rewrote the GitHub-facing README to work better as a landing page, with a sharper value prop, clearer install guidance, a real quickstart, example requests, a minimal config example, troubleshooting, and a high-level flow diagram
+- README now explains when to use `pi-oracle`, when not to use it first, where outputs live, and what to do if a wake-up is missed
+
+### Fixed
+- removed the stale legacy model wording from the README example flow and aligned user-facing setup docs with the preset-based configuration model
+
+## 0.3.0 - 2026-04-08
+
+### Changed
+- breaking: `oracle_submit` and oracle config defaults now use preset-only model selection; legacy `modelFamily` / `effort` / `autoSwitchToThinking` submit inputs and default config fields were removed in favor of canonical preset ids
+- oracle jobs now persist a resolved `selection` snapshot and the worker configures ChatGPT from that persisted selection instead of re-deriving model settings from legacy job fields
+- oracle model preset definitions now come from a single canonical registry in `extensions/oracle/lib/config.ts`
+
+### Fixed
+- removed duplicate hand-maintained preset-id examples from agent-facing prompt and design docs so callers are directed to the tool schema / canonical registry instead of stale inline lists
+- oracle sanity coverage now validates the preset-only contract from the registered tool schema and canonical registry instead of brittle prose-only assertions
+- worker model configuration now consistently uses the explicit `configureModel(job)` parameter instead of hidden coupling through the module-global current job
+
 ## 0.2.2 - 2026-04-07
 
 ### Fixed

package/README.md

@@ -1,152 +1,191 @@
 # pi-oracle
 
-`pi-oracle` is a `pi` extension that lets the agent use ChatGPT.com as a long-running web oracle instead of using the API.
+`pi-oracle` is a `pi` package that lets the agent hand off difficult, long-running tasks to ChatGPT.com through the web app instead of the API.
 
-It exists for the hard cases where you want:
-- the user’s real ChatGPT account
+Use it when you want:
+- your real ChatGPT account
 - web-model behavior instead of API usage
-- large project-context uploads
-- async background execution with durable job-state/response artifacts plus best-effort wake-ups for the originating `pi` session
-- oracle requires a persisted `pi` session identity; in-memory/no-session contexts are rejected instead of risking wrong-session wake-ups
-- legacy project-scoped jobs created before that change remain readable via project status/read commands, but skip best-effort wake-up routing after upgrade
+- large repo/context uploads
+- async background execution
+- durable saved responses/artifacts plus best-effort wake-ups back into `pi`
 
-Normal oracle jobs run in an isolated browser profile, not in the user’s active Chrome window.
+Normal oracle jobs run in an isolated browser profile, not your active Chrome window.
 
-Status: experimental public beta, validated primarily on macOS.
+> Status: experimental public beta. Validated primarily on macOS with Google Chrome and `pi` 0.65.0+.
 
-Compatibility target: current lifecycle/event model in `pi` 0.65.0+.
-This package intentionally uses the current `session_start`-based session lifecycle and does not ship backward-compatibility shims for removed extension events.
+## When to use it
 
-## What it does
+Use `pi-oracle` for:
+- big code reviews of a repo or pending changes
+- architectural or migration analysis that benefits from a large uploaded archive
+- long-running prompts that may take minutes to finish
+- follow-up questions in the same ChatGPT thread later
 
-The extension adds:
-- `/oracle <request>`
-  - implemented as a prompt template so it keeps native pi queueing behavior during streaming and compaction
-- `/oracle-auth`
-- `/oracle-status [job-id]`
-- `/oracle-cancel [job-id]`
-- `/oracle-clean <job-id|all>`
+Do not reach for it first for:
+- normal short coding tasks that `pi` can handle directly
+- workflows that must never upload project archives to ChatGPT.com
+- environments outside the current supported setup
+
+## Install
+
+npm:
+
+```bash
+pi install npm:pi-oracle
+```
+
+GitHub:
+
+```bash
+pi install https://github.com/fitchmultz/pi-oracle
+```
+
+## Quickstart
+
+1. Make sure ChatGPT already works in your local Chrome profile.
+2. Make sure these are installed: Google Chrome, `agent-browser`, `tar`, and `zstd`.
+3. Optional: create `~/.pi/agent/extensions/oracle.json` if you want non-default settings.
+4. Run `/oracle-auth`.
+5. Run `/oracle Review the current pending changes. Include the whole repo unless a narrower archive is clearly better.`
+6. Wait for a best-effort wake-up, or check `/oracle-status`.
+
+If you miss the wake-up, the result is still saved durably in the oracle job directory and can be read later.
+
+## Example requests
+
+```text
+/oracle Review the current pending changes. Include the whole repo unless a narrower archive is clearly better. Give me a prioritized code review with concrete fixes.
+```
+
+```text
+/oracle Read the codebase and explain the highest-risk auth/session failure modes, including what to test before shipping.
+```
+
+## High-level flow
+
+```mermaid
+flowchart LR
+    A["/oracle request"] --> B["Agent gathers repo context"]
+    B --> C["oracle_submit builds archive"]
+    C --> D["Detached worker starts isolated ChatGPT runtime"]
+    D --> E["Archive + prompt sent to ChatGPT.com"]
+    E --> F["Response/artifacts saved under oracle job dir"]
+    F --> G["Best-effort wake-up to matching pi session"]
+```
+
+If concurrency is full, the job is queued and starts automatically later.
+
+## What the package adds
+
+User-facing commands:
+- `/oracle <request>` — prompt template that tells the agent to gather context and dispatch an oracle job
+- `/oracle-auth` — sync ChatGPT cookies from your real Chrome profile into the isolated oracle auth profile
+- `/oracle-status [job-id]` — inspect job status
+- `/oracle-cancel [job-id]` — cancel queued or active job
+- `/oracle-clean <job-id|all>` — remove temp files for terminal jobs
+
+Agent-facing tools:
 - `oracle_submit`
 - `oracle_read`
 - `oracle_cancel`
 
-An oracle job:
-1. gathers a project archive
-2. if runtime capacity is full, persists as `queued` and starts automatically later
-3. otherwise opens ChatGPT in an isolated runtime profile
-4. uploads the archive and sends the prompt
-5. waits in the background
-6. persists the response and any artifacts under the oracle job directory (`${PI_ORACLE_JOBS_DIR:-/tmp}/oracle-<job-id>/` by default)
-   - old terminal jobs are later pruned according to cleanup retention settings
-   - when directory inputs are expanded, project archives automatically skip common bulky generated caches and top-level build outputs such as `node_modules/`, `target/`, virtualenv caches, coverage outputs, and `dist/`/`build/`/`out/`, unless you explicitly pass those directories
-   - whole-repo archive defaults also skip obvious credentials/private data such as `.env` files, key material, credential dotfiles, local database files, and root `secrets/` directories unless you explicitly pass them
-   - if a whole-repo archive is still too large after default exclusions, submit automatically prunes the largest nested directories with generic generated-output names like `build/`, `dist/`, `out/`, `coverage/`, and `tmp/` outside obvious source roots like `src/` and `lib/`, and successful submissions report what was pruned
-7. persists the response/artifacts durably in oracle job state and issues best-effort wake-ups to whichever matching `pi` session is currently live
-
-## Example
+## Minimal config
 
-```text
-/oracle Invoke the Oracle to have it generate a thorough code review of the current pending changes. By default include the whole repo archive unless the request clearly needs a narrower scope. Use the Pro Model with Extended effort.
+Most users can start with the packaged defaults and only set the Chrome profile if needed.
+
+`~/.pi/agent/extensions/oracle.json`
+
+```json
+{
+  "defaults": {
+    "preset": "<preset id from ORACLE_SUBMIT_PRESETS>"
+  },
+  "auth": {
+    "chromeProfile": "Default"
+  }
+}
 ```
 
-## Why this exists
+Notes:
+- `defaults.preset` is the default ChatGPT model preset for oracle jobs.
+- The canonical preset ids live in `extensions/oracle/lib/config.ts`.
+- If the packaged default is fine, you can omit `defaults.preset` entirely.
+- You usually do not need to set browser paths unless auto-detection fails.
+
+Other useful settings:
+- `browser.runMode`
+- `browser.args`
+- `browser.authSeedProfileDir`
+- `browser.runtimeProfilesDir`
+- `cleanup.completeJobRetentionMs`
+- `cleanup.failedJobRetentionMs`
 
-The goal is to get strong ChatGPT web-model answers without:
-- paying API costs for every long review
-- blocking the agent for 10–90 minutes
-- stealing focus from the user’s active browser session
+Project config should only override safe, non-privileged settings.
 
-## Current scope
+## What happens to outputs
 
-Currently validated for:
-- macOS
-- local Google Chrome
-- local ChatGPT web login in Chrome
-- isolated auth seed profile + per-job runtime profile clones
-- concurrent jobs across different projects/sessions
-- workerless queued jobs when the global concurrency limit is full
-- same-conversation exclusion for follow-ups
-- plain-text responses
-- artifact capture, including multi-artifact runs
-
-Not promised yet:
-- cross-platform support
-- immunity to future ChatGPT UI changes
-- fully polished partial-artifact terminal semantics
+- Jobs persist their response and any artifacts under `${PI_ORACLE_JOBS_DIR:-/tmp}/oracle-<job-id>/` by default.
+- Jobs can queue automatically if runtime capacity is full.
+- Completion delivery into `pi` is best-effort wake-up based.
+- If you miss the wake-up, use `oracle_read(jobId)` or `/oracle-status`.
 
 ## Requirements
 
 - macOS
 - Google Chrome installed
 - ChatGPT already signed into a local Chrome profile
-- `pi` 0.65.0 or newer installed
+- `pi` 0.65.0 or newer
 - `agent-browser` available on the machine
 - `tar` and `zstd` available
 
-## Install
+## Troubleshooting
 
-npm:
+### `/oracle-auth` fails or says login is required
 
-```bash
-pi install npm:pi-oracle
-```
+- Make sure ChatGPT works in the same local Chrome profile you configured.
+- Re-run `/oracle-auth`.
+- If ChatGPT is half-logged-in or challenge flow state looks weird, finish the login/challenge in the headed auth browser and retry.
 
-GitHub:
+### You hit a challenge / verification page
 
-```bash
-pi install https://github.com/fitchmultz/pi-oracle
-```
+- Solve it in the auth/bootstrap browser if prompted.
+- Then re-run `/oracle-auth` before submitting jobs again.
 
-## First-time setup
+### You see "Oracle requires a persisted pi session"
 
-1. Make sure ChatGPT already works in your local Chrome profile.
-2. Configure the oracle if needed via `~/.pi/agent/extensions/oracle.json`.
-3. Run `/oracle-auth`.
-4. Run a tiny `/oracle` smoke test.
+- Do not run oracle from `pi --no-session`.
+- Start a normal persisted `pi` session, then use `/oracle` again.
 
-## Configuration
+### A job finished but no wake-up arrived
 
-Config files:
-- global: `~/.pi/agent/extensions/oracle.json`
-- project: `.pi/extensions/oracle.json`
+- Use `/oracle-status [job-id]` or `oracle_read(jobId)`.
+- Results are still saved on disk even if the reminder turn does not land.
 
-Common settings:
-- `browser.args`
-- `browser.executablePath`
-- `browser.authSeedProfileDir`
-- `browser.runtimeProfilesDir`
-- `auth.chromeProfile`
-- `auth.chromeCookiePath`
-- `cleanup.completeJobRetentionMs`
-- `cleanup.failedJobRetentionMs`
+### `agent-browser`, `tar`, or `zstd` is missing
 
-Project config should only override safe, non-privileged settings.
+- Install the missing local dependency and rerun the command.
+
+### Auto-detection picked the wrong Chrome profile
+
+- Set `auth.chromeProfile` in `~/.pi/agent/extensions/oracle.json`.
+- Re-run `/oracle-auth`.
+
+### You want more details about a failed run
+
+- Inspect the job directory under `${PI_ORACLE_JOBS_DIR:-/tmp}/oracle-<job-id>/`.
+- The worker log and captured diagnostics are stored there.
+
+## Detailed docs
 
-Cleanup behavior:
-- terminal job directories under the configured oracle jobs dir (`${PI_ORACLE_JOBS_DIR:-/tmp}/oracle-<job-id>/` by default) are retained for inspection, then pruned conservatively
-- completed/cancelled jobs are pruned after `cleanup.completeJobRetentionMs` based on terminal-job age, but recent wake-up sends keep response/artifact files retained briefly so follow-up turns do not point at deleted paths
-- failed jobs are pruned after `cleanup.failedJobRetentionMs`
-- `/oracle-cancel` can cancel queued or active jobs
-- `/oracle-clean` refuses non-terminal jobs, including queued ones, refuses terminal jobs whose worker is still live, also refuses recently woken jobs during a short post-send retention grace window, performs runtime cleanup before removing terminal job directories, and reports cleanup warnings if any residual cleanup step fails
-
-Detailed design and maintainer docs:
-- `docs/ORACLE_DESIGN.md`
-- `docs/ORACLE_RECOVERY_DRILL.md`
-
-Completion notification semantics:
-- oracle responses and artifacts are always persisted durably in oracle job state under the configured oracle jobs dir (`${PI_ORACLE_JOBS_DIR:-/tmp}/oracle-<job-id>/` by default)
-- completion delivery into pi sessions is best-effort wake-up based; the extension no longer appends synthetic assistant completion messages into session history
-- wake-up content tells the receiving assistant to call `oracle_read(jobId)` as the canonical completion-consumption path, with saved file paths included as secondary context
-- manual `oracle_read` or `/oracle-status` inspection settles further reminder retries once the terminal job has been opened, and now persists settlement provenance for postmortems
-- manual inspection before the first wake-up attempt is recorded as observation only and does not suppress the first reminder send
-- if a wake-up does not land, the job remains available via its saved response/artifacts and status commands
+- `docs/ORACLE_DESIGN.md` — architecture, lifecycle, queueing, persistence, presets, and recovery behavior
+- `docs/ORACLE_RECOVERY_DRILL.md` — safe expired-auth recovery validation drill
 
 ## Privacy / local data
 
 This extension is local-first, but it does read and persist local data:
 - `/oracle-auth` reads ChatGPT cookies from a local Chrome profile
 - job archives are uploaded to ChatGPT.com
-- responses and artifacts are written under the configured oracle jobs dir (`${PI_ORACLE_JOBS_DIR:-/tmp}/oracle-<job-id>/` by default)
+- responses and artifacts are written under the configured oracle jobs dir
 
 Review the code and design docs before using it with sensitive material.
 

package/docs/ORACLE_DESIGN.md

@@ -133,7 +133,9 @@ The authenticated seed profile remains the source of truth for future oracle run
 
 ### `oracle_submit`
 
-1. validate config and model options
+Agent-facing submissions use **`preset`**; the canonical registry is `ORACLE_SUBMIT_PRESETS` in `extensions/oracle/lib/config.ts`. **`preset` is the only model-selection parameter** on `oracle_submit`. There are no `modelFamily`, `effort`, or `autoSwitchToThinking` fields.
+
+1. resolve the preset (submit-time or config default) into an execution snapshot
 2. resolve optional `followUpJobId` into a prior `chatUrl` and `conversationId`
 3. build the archive first into a temporary path
 4. allocate a unique runtime:
@@ -216,9 +218,7 @@ Browser/auth settings are global-only because they control local privileged brow
 ```json
 {
   "defaults": {
-    "modelFamily": "pro",
-    "effort": "extended",
-    "autoSwitchToThinking": false
+    "preset": "<preset id from ORACLE_SUBMIT_PRESETS>"
   },
   "browser": {
     "sessionPrefix": "oracle",
@@ -317,9 +317,7 @@ Important fields include:
 - `sessionId`
 - `originSessionFile`
 - `requestSource`
-- `chatModelFamily`
-- `effort`
-- `autoSwitchToThinking`
+- `selection`: resolved execution snapshot with `{ preset, modelFamily, effort?, autoSwitchToThinking }`
 - `followUpToJobId`
 - `chatUrl`
 - `conversationId`

package/extensions/oracle/lib/config.ts

@@ -10,13 +10,61 @@ export type OracleModelFamily = (typeof MODEL_FAMILIES)[number];
 export const EFFORTS = ["light", "standard", "extended", "heavy"] as const;
 export type OracleEffort = (typeof EFFORTS)[number];
 
+/**
+ * Canonical preset registry for `oracle_submit` preset selection.
+ * This is the single authored source of truth — all derived lists come from `Object.keys(...)`.
+ */
+export const ORACLE_SUBMIT_PRESETS = {
+  pro_standard: { label: "Pro - Standard", modelFamily: "pro" as const, effort: "standard" as const, autoSwitchToThinking: false },
+  pro_extended: { label: "Pro - Extended", modelFamily: "pro" as const, effort: "extended" as const, autoSwitchToThinking: false },
+  thinking_light: { label: "Thinking - Light", modelFamily: "thinking" as const, effort: "light" as const, autoSwitchToThinking: false },
+  thinking_standard: { label: "Thinking - Standard", modelFamily: "thinking" as const, effort: "standard" as const, autoSwitchToThinking: false },
+  thinking_extended: { label: "Thinking - Extended", modelFamily: "thinking" as const, effort: "extended" as const, autoSwitchToThinking: false },
+  thinking_heavy: { label: "Thinking - Heavy", modelFamily: "thinking" as const, effort: "heavy" as const, autoSwitchToThinking: false },
+  instant: { label: "Instant", modelFamily: "instant" as const, autoSwitchToThinking: false },
+  instant_auto_switch: { label: "Instant - Auto-switch to Thinking Enabled", modelFamily: "instant" as const, autoSwitchToThinking: true },
+} as const;
+
+export type OracleSubmitPresetId = keyof typeof ORACLE_SUBMIT_PRESETS;
+
+export type OracleSubmitPreset = typeof ORACLE_SUBMIT_PRESETS[OracleSubmitPresetId];
+
+export function getOracleSubmitPresetById(id: OracleSubmitPresetId): OracleSubmitPreset {
+  const found = ORACLE_SUBMIT_PRESETS[id];
+  if (!found) {
+    throw new Error(`Unknown oracle_submit preset: ${id}`);
+  }
+  return found;
+}
+
+/** Resolved execution snapshot generated from a preset at submit time. */
+export type OracleResolvedSelection = {
+  preset: OracleSubmitPresetId;
+  modelFamily: OracleModelFamily;
+  effort?: OracleEffort;
+  autoSwitchToThinking: boolean;
+};
+
+/**
+ * Resolve a preset id into the execution snapshot that gets persisted on the job.
+ * @throws if the preset id is unknown.
+ */
+export function resolveOracleSubmitPreset(presetId: OracleSubmitPresetId): OracleResolvedSelection {
+  const def = getOracleSubmitPresetById(presetId);
+  return {
+    preset: presetId,
+    modelFamily: def.modelFamily,
+    effort: def.modelFamily === "instant" ? undefined : def.effort,
+    autoSwitchToThinking: def.modelFamily === "instant" ? def.autoSwitchToThinking : false,
+  };
+}
+
 export const BROWSER_RUN_MODES = ["headless", "headed"] as const;
 export type OracleBrowserRunMode = (typeof BROWSER_RUN_MODES)[number];
 
 export const CLONE_STRATEGIES = ["apfs-clone", "copy"] as const;
 export type OracleCloneStrategy = (typeof CLONE_STRATEGIES)[number];
 
-const PRO_EFFORTS = ["standard", "extended"] as const satisfies readonly OracleEffort[];
 const ALLOWED_CHATGPT_ORIGINS = new Set(["https://chatgpt.com", "https://chat.openai.com"]);
 const PROJECT_OVERRIDE_KEYS = new Set(["defaults", "worker", "poller", "artifacts", "cleanup"]);
 const DEFAULT_MAC_CHROME_EXECUTABLE = "/Applications/Google Chrome.app/Contents/MacOS/Google Chrome";
@@ -24,9 +72,7 @@ const DEFAULT_MAC_CHROME_USER_DATA_DIR = join(homedir(), "Library", "Application
 
 export interface OracleConfig {
   defaults: {
-    modelFamily: OracleModelFamily;
-    effort: OracleEffort;
-    autoSwitchToThinking: boolean;
+    preset: OracleSubmitPresetId;
   };
   browser: {
     sessionPrefix: string;
@@ -98,9 +144,7 @@ const detectedChromeProfileName = detectDefaultChromeProfileName();
 
 export const DEFAULT_CONFIG: OracleConfig = {
   defaults: {
-    modelFamily: "pro",
-    effort: "extended",
-    autoSwitchToThinking: false,
+    preset: "pro_extended",
   },
   browser: {
     sessionPrefix: "oracle",
@@ -292,19 +336,13 @@ function normalizeLegacyBrowserConfig(root: Record<string, unknown>): Record<str
   return root;
 }
 
+const PRESET_IDS = Object.keys(ORACLE_SUBMIT_PRESETS) as unknown as readonly OracleSubmitPresetId[];
+
 function validateOracleConfig(value: unknown): OracleConfig {
   const root = normalizeLegacyBrowserConfig(expectObject(value, "root"));
 
   const defaults = expectObject(root.defaults, "defaults");
-  const modelFamily = expectEnum(defaults.modelFamily, "defaults.modelFamily", MODEL_FAMILIES);
-  const effort = expectEnum(defaults.effort, "defaults.effort", EFFORTS);
-  const autoSwitchToThinking = expectBoolean(defaults.autoSwitchToThinking, "defaults.autoSwitchToThinking");
-  if (modelFamily === "pro" && effort !== "standard" && effort !== "extended") {
-    throw new Error(`Invalid oracle config: defaults.effort must be one of ${PRO_EFFORTS.join(", ")} for pro`);
-  }
-  if (modelFamily !== "instant" && autoSwitchToThinking) {
-    throw new Error("Invalid oracle config: defaults.autoSwitchToThinking is only valid for instant");
-  }
+  const preset = expectEnum(defaults.preset, "defaults.preset", PRESET_IDS);
 
   const browser = expectObject(root.browser, "browser");
   const auth = expectObject(root.auth, "auth");
@@ -321,9 +359,7 @@ function validateOracleConfig(value: unknown): OracleConfig {
 
   return {
     defaults: {
-      modelFamily,
-      effort,
-      autoSwitchToThinking,
+      preset,
     },
     browser: {
       sessionPrefix: expectString(browser.sessionPrefix, "browser.sessionPrefix"),

package/extensions/oracle/lib/jobs.ts

@@ -4,7 +4,7 @@ import { existsSync, readdirSync, readFileSync } from "node:fs";
 import { chmod, mkdir, readFile, rename, rm, writeFile } from "node:fs/promises";
 import { join, resolve } from "node:path";
 import type { ExtensionContext } from "@mariozechner/pi-coding-agent";
-import type { OracleConfig, OracleEffort, OracleModelFamily } from "./config.js";
+import type { OracleConfig, OracleResolvedSelection } from "./config.js";
 import { withJobLock, withLock } from "./locks.js";
 import { cleanupRuntimeArtifacts, getProjectId, getSessionId, parseConversationId, requirePersistedSessionFile, type OracleCleanupReport } from "./runtime.js";
 
@@ -133,9 +133,7 @@ export interface OracleJob {
   sessionId: string;
   originSessionFile?: string;
   requestSource: "command" | "tool";
-  chatModelFamily: OracleModelFamily;
-  effort?: OracleEffort;
-  autoSwitchToThinking?: boolean;
+  selection: OracleResolvedSelection;
   followUpToJobId?: string;
   chatUrl?: string;
   conversationId?: string;
@@ -185,9 +183,7 @@ export interface OracleJob {
 export interface OracleSubmitInput {
   prompt: string;
   files: string[];
-  modelFamily: OracleModelFamily;
-  effort?: OracleEffort;
-  autoSwitchToThinking?: boolean;
+  selection: OracleResolvedSelection;
   followUpToJobId?: string;
   chatUrl?: string;
   requestSource: "command" | "tool";
@@ -971,10 +967,6 @@ export async function createJob(
 
   const createdAt = options?.createdAt ?? new Date().toISOString();
   const initialState = options?.initialState ?? "submitted";
-  const normalizedEffort = input.modelFamily === "instant" ? undefined : (input.effort ?? config.defaults.effort);
-  const normalizedAutoSwitchToThinking = input.modelFamily === "instant"
-    ? (input.autoSwitchToThinking ?? config.defaults.autoSwitchToThinking)
-    : false;
   const job: OracleJob = {
     id,
     status: initialState,
@@ -988,9 +980,7 @@ export async function createJob(
     sessionId,
     originSessionFile: sessionFile,
     requestSource: input.requestSource,
-    chatModelFamily: input.modelFamily,
-    effort: normalizedEffort,
-    autoSwitchToThinking: normalizedAutoSwitchToThinking,
+    selection: input.selection,
     followUpToJobId: input.followUpToJobId,
     chatUrl: input.followUpToJobId ? input.chatUrl : undefined,
     conversationId,

package/extensions/oracle/lib/tools.ts

@@ -5,7 +5,12 @@ import { basename, join, posix } from "node:path";
 import type { ExtensionAPI } from "@mariozechner/pi-coding-agent";
 import { Type } from "@sinclair/typebox";
 import { isLockTimeoutError, withGlobalReconcileLock, withLock } from "./locks.js";
-import { loadOracleConfig, EFFORTS, MODEL_FAMILIES, type OracleEffort, type OracleModelFamily } from "./config.js";
+import {
+  loadOracleConfig,
+  ORACLE_SUBMIT_PRESETS,
+  resolveOracleSubmitPreset,
+  type OracleSubmitPresetId,
+} from "./config.js";
 import {
   appendCleanupWarnings,
   cancelOracleJob,
@@ -53,10 +58,11 @@ const ORACLE_SUBMIT_PARAMS = Type.Object({
     description: "Exact project-relative files/directories to include in the oracle archive.",
     minItems: 1,
   }),
-  modelFamily: Type.Optional(stringEnum(MODEL_FAMILIES, "ChatGPT model family: instant, thinking, or pro.")),
-  effort: Type.Optional(stringEnum(EFFORTS, "Reasoning effort. Use only values supported by the chosen model family.")),
-  autoSwitchToThinking: Type.Optional(
-    Type.Boolean({ description: "Only valid when modelFamily is instant. Omit for thinking and pro." }),
+  preset: Type.Optional(
+    stringEnum(
+      [...Object.keys(ORACLE_SUBMIT_PRESETS)] as const,
+      "ChatGPT model preset. Omit to use the configured default preset.",
+    ),
   ),
   followUpJobId: Type.Optional(Type.String({ description: "Earlier oracle job id whose chat thread should be continued." })),
 });
@@ -69,12 +75,6 @@ const ORACLE_CANCEL_PARAMS = Type.Object({
   jobId: Type.String({ description: "Oracle job id." }),
 });
 
-const VALID_EFFORTS: Record<OracleModelFamily, readonly OracleEffort[]> = {
-  instant: [],
-  thinking: ["light", "standard", "extended", "heavy"],
-  pro: ["standard", "extended"],
-};
-
 const MAX_ARCHIVE_BYTES = 250 * 1024 * 1024;
 const MAX_QUEUED_JOBS_PER_ACTIVE_RUNTIME = 1;
 const MAX_QUEUED_ARCHIVE_BYTES_PER_ACTIVE_RUNTIME = MAX_ARCHIVE_BYTES;
@@ -491,29 +491,6 @@ export function getQueueAdmissionFailure(args: {
   return undefined;
 }
 
-function validateSubmissionOptions(
-  params: { effort?: OracleEffort; autoSwitchToThinking?: boolean },
-  modelFamily: OracleModelFamily,
-  effort: OracleEffort | undefined,
-  autoSwitchToThinking: boolean,
-): void {
-  if (modelFamily === "instant" && params.effort !== undefined) {
-    throw new Error("Instant model family does not support effort selection");
-  }
-
-  if (effort && !VALID_EFFORTS[modelFamily].includes(effort)) {
-    throw new Error(`Invalid effort for ${modelFamily}: ${effort}`);
-  }
-
-  if (modelFamily !== "instant" && params.autoSwitchToThinking === true) {
-    throw new Error("autoSwitchToThinking is only valid for the instant model family");
-  }
-
-  if (modelFamily !== "instant" && autoSwitchToThinking) {
-    throw new Error(`autoSwitchToThinking cannot be enabled for ${modelFamily}`);
-  }
-}
-
 function resolveFollowUp(previousJobId: string | undefined, cwd: string): {
   followUpToJobId?: string;
   chatUrl?: string;
@@ -601,7 +578,8 @@ export function registerOracleTools(pi: ExtensionAPI, workerPath: string): void
     name: "oracle_submit",
     label: "Oracle Submit",
     description:
-      "Dispatch a background ChatGPT web oracle job after gathering context. Always pass a prompt and exact project-relative archive inputs.",
+      "Dispatch a background ChatGPT web oracle job after gathering context. Always pass a prompt and exact project-relative archive inputs. " +
+      "Optional ChatGPT model: set parameter `preset`, or omit it for configured defaults (see `preset` field for allowed ids).",
     promptSnippet: "Dispatch a background ChatGPT web oracle job after gathering repo context.",
     promptGuidelines: [
       "Gather context before calling oracle_submit.",
@@ -613,7 +591,7 @@ export function registerOracleTools(pi: ExtensionAPI, workerPath: string): void
       "If oracle_submit itself fails because the local archive still exceeds the upload limit after default exclusions and automatic generic generated-output-dir pruning, or for any other submit-time error, stop and report the error instead of retrying automatically.",
       "If oracle_submit returns a queued job instead of an immediately dispatched one, treat that as success and stop exactly the same way.",
       "Stop after dispatching oracle_submit; do not continue the task while the oracle job is running.",
-      "Only use autoSwitchToThinking with modelFamily=instant.",
+      "Use `preset` as the only model-selection parameter on oracle_submit. Allowed values come from the tool schema enum. Omit preset to use the configured default.",
     ],
     parameters: ORACLE_SUBMIT_PARAMS,
     async execute(_toolCallId, params, _signal, _onUpdate, ctx) {
@@ -621,16 +599,9 @@ export function registerOracleTools(pi: ExtensionAPI, workerPath: string): void
       const originSessionFile = requirePersistedSessionFile(getSessionFile(ctx), "submit oracle jobs");
       const projectId = getProjectId(ctx.cwd);
       const sessionId = getSessionId(originSessionFile, projectId);
-      const submittedModelFamily = params.modelFamily as OracleModelFamily | undefined;
-      const submittedEffort = params.effort as OracleEffort | undefined;
-      const modelFamily: OracleModelFamily = submittedModelFamily ?? config.defaults.modelFamily;
-      const requestedEffort: OracleEffort = submittedEffort ?? config.defaults.effort;
-      const effort: OracleEffort | undefined = modelFamily === "instant" ? undefined : requestedEffort;
-      const rawAutoSwitchToThinking = params.autoSwitchToThinking ?? config.defaults.autoSwitchToThinking;
-      const autoSwitchToThinking = modelFamily === "instant" ? rawAutoSwitchToThinking : false;
+      const presetId = (params.preset as OracleSubmitPresetId | undefined) ?? config.defaults.preset;
+      const selection = resolveOracleSubmitPreset(presetId);
       const followUp = resolveFollowUp(params.followUpJobId, ctx.cwd);
-
-      validateSubmissionOptions({ effort: submittedEffort, autoSwitchToThinking: params.autoSwitchToThinking }, modelFamily, effort, autoSwitchToThinking);
       try {
         await withGlobalReconcileLock({ processPid: process.pid, source: "oracle_submit", cwd: ctx.cwd }, async () => {
           await reconcileStaleOracleJobs();
@@ -691,9 +662,7 @@ export function registerOracleTools(pi: ExtensionAPI, workerPath: string): void
               {
                 prompt: params.prompt,
                 files: params.files,
-                modelFamily,
-                effort,
-                autoSwitchToThinking,
+                selection,
                 followUpToJobId: followUp.followUpToJobId,
                 chatUrl: followUp.chatUrl,
                 requestSource: "tool",
@@ -736,9 +705,7 @@ export function registerOracleTools(pi: ExtensionAPI, workerPath: string): void
             {
               prompt: params.prompt,
               files: params.files,
-              modelFamily,
-              effort,
-              autoSwitchToThinking,
+              selection,
               followUpToJobId: followUp.followUpToJobId,
               chatUrl: followUp.chatUrl,
               requestSource: "tool",

package/extensions/oracle/worker/run-job.mjs

@@ -807,7 +807,7 @@ function matchesModelFamilyButton(candidate, family) {
 }
 
 function requestedEffortLabel(job) {
-  return job.effort ? titleCase(job.effort) : undefined;
+  return job.selection?.effort ? titleCase(job.selection.effort) : undefined;
 }
 
 function effortSelectionVisible(snapshot, effortLabel) {
@@ -852,12 +852,12 @@ function snapshotHasModelConfigurationUi(snapshot) {
 
 function snapshotStronglyMatchesRequestedModel(snapshot, job) {
   const entries = parseSnapshotEntries(snapshot);
-  const familyMatched = entries.some((entry) => matchesModelFamilyButton(entry, job.chatModelFamily));
+  const familyMatched = entries.some((entry) => matchesModelFamilyButton(entry, job.selection.modelFamily));
   const effortLabel = requestedEffortLabel(job);
-  if (job.chatModelFamily === "thinking") {
+  if (job.selection.modelFamily === "thinking") {
     return familyMatched || effortSelectionVisible(snapshot, effortLabel);
   }
-  if (job.chatModelFamily === "pro") {
+  if (job.selection.modelFamily === "pro") {
     return effortLabel ? familyMatched && effortSelectionVisible(snapshot, effortLabel) : familyMatched;
   }
   return familyMatched;
@@ -880,13 +880,13 @@ function composerControlsVisible(snapshot) {
 }
 
 function snapshotWeaklyMatchesRequestedModel(snapshot, job) {
-  if (job.chatModelFamily === "thinking") {
+  if (job.selection.modelFamily === "thinking") {
     return effortSelectionVisible(snapshot, requestedEffortLabel(job)) || thinkingSelectionVisible(snapshot);
   }
-  if (job.chatModelFamily === "pro") {
+  if (job.selection.modelFamily === "pro") {
     return !thinkingChipVisible(snapshot);
   }
-  if (job.chatModelFamily === "instant") {
+  if (job.selection.modelFamily === "instant") {
     return !thinkingChipVisible(snapshot);
   }
   return false;
@@ -1214,7 +1214,7 @@ async function waitForModelConfigurationToSettle(job, options = {}) {
       if (options.stronglyVerified) {
         if (!fallbackLogged) {
           fallbackLogged = true;
-          await log(`Model configuration closed after strong in-dialog verification for family=${job.chatModelFamily} effort=${job.effort || "(none)"}`);
+          await log(`Model configuration closed after strong in-dialog verification for family=${job.selection.modelFamily} effort=${job.selection?.effort || "(none)"}`);
         }
         return;
       }
@@ -1223,7 +1223,7 @@ async function waitForModelConfigurationToSettle(job, options = {}) {
     if (!configurationUiVisible && composerControlsVisible(snapshot) && options.stronglyVerified) {
       if (!fallbackLogged) {
         fallbackLogged = true;
-        await log(`Composer became usable after strong in-dialog verification for family=${job.chatModelFamily} effort=${job.effort || "(none)"}`);
+        await log(`Composer became usable after strong in-dialog verification for family=${job.selection.modelFamily} effort=${job.selection?.effort || "(none)"}`);
       }
       return;
     }
@@ -1239,30 +1239,30 @@ async function waitForModelConfigurationToSettle(job, options = {}) {
   }
 
   if (options.stronglyVerified && lastSnapshot && !snapshotHasModelConfigurationUi(lastSnapshot)) {
-    await log(`Model configuration closed only after settle-timeout for family=${job.chatModelFamily} effort=${job.effort || "(none)"}`);
+    await log(`Model configuration closed only after settle-timeout for family=${job.selection.modelFamily} effort=${job.selection?.effort || "(none)"}`);
     return;
   }
 
-  throw new Error(`Could not verify requested model settings after configuration for ${job.chatModelFamily}`);
+  throw new Error(`Could not verify requested model settings after configuration for ${job.selection.modelFamily}`);
 }
 
 async function configureModel(job) {
   const initialSnapshot = await snapshotText(job);
   if (snapshotStronglyMatchesRequestedModel(initialSnapshot, job)) {
-    await log(`Model already appears configured for family=${job.chatModelFamily} effort=${job.effort || "(none)"}; skipping reconfiguration`);
+    await log(`Model already appears configured for family=${job.selection.modelFamily} effort=${job.selection?.effort || "(none)"}; skipping reconfiguration`);
     return;
   }
 
-  await log(`Configuring model family=${job.chatModelFamily} effort=${job.effort || "(none)"}`);
+  await log(`Configuring model family=${job.selection.modelFamily} effort=${job.selection?.effort || "(none)"}`);
   let familySnapshot = await openModelConfiguration(job);
   let verificationSnapshot = familySnapshot;
 
-  let familyEntry = findEntry(familySnapshot, (candidate) => matchesModelFamilyButton(candidate, job.chatModelFamily));
+  let familyEntry = findEntry(familySnapshot, (candidate) => matchesModelFamilyButton(candidate, job.selection.modelFamily));
   if (!familyEntry && snapshotStronglyMatchesRequestedModel(familySnapshot, job)) {
     await log("Model configuration UI opened with requested settings already selected");
   }
   if (!familyEntry && !snapshotStronglyMatchesRequestedModel(familySnapshot, job)) {
-    throw new Error(`Could not find model family button for ${job.chatModelFamily}`);
+    throw new Error(`Could not find model family button for ${job.selection.modelFamily}`);
   }
 
   if (familyEntry) {
@@ -1272,7 +1272,7 @@ async function configureModel(job) {
     verificationSnapshot = familySnapshot;
   }
 
-  if (job.chatModelFamily === "thinking" || job.chatModelFamily === "pro") {
+  if (job.selection.modelFamily === "thinking" || job.selection.modelFamily === "pro") {
     const effortLabel = requestedEffortLabel(job);
     if (effortLabel && !effortSelectionVisible(familySnapshot, effortLabel)) {
       const opened = await openEffortDropdown(job);
@@ -1294,14 +1294,14 @@ async function configureModel(job) {
     }
   }
 
-  if (job.chatModelFamily === "instant" && job.autoSwitchToThinking) {
+  if (job.selection.modelFamily === "instant" && job.selection.autoSwitchToThinking) {
     await maybeClickLabeledEntry(job, CHATGPT_LABELS.autoSwitchToThinking);
     verificationSnapshot = await snapshotText(job);
   }
 
   const stronglyVerified = snapshotStronglyMatchesRequestedModel(verificationSnapshot, job);
   if (!stronglyVerified) {
-    throw new Error(`Could not verify requested model settings in configuration UI for ${job.chatModelFamily}`);
+    throw new Error(`Could not verify requested model settings in configuration UI for ${job.selection.modelFamily}`);
   }
 
   if (!(await maybeClickLabeledEntry(job, CHATGPT_LABELS.close, { kind: "button" }))) {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "pi-oracle",
-  "version": "0.2.2",
+  "version": "0.3.1",
   "description": "ChatGPT web-oracle extension for pi with isolated browser auth, async jobs, and project-context archives.",
   "private": false,
   "license": "MIT",

package/prompts/oracle.md

@@ -13,6 +13,12 @@ Required workflow:
 5. Call oracle_submit with the prompt and exact archive inputs.
 6. Stop immediately after dispatching the oracle job.
 
+Oracle model (`oracle_submit`):
+- To choose a specific ChatGPT model, pass **`preset`** with one of the allowed ids from the tool schema enum / canonical preset registry.
+- **Or** omit **`preset`** entirely to use the configured default model (from oracle config).
+- **`preset`** is the only model-selection parameter on `oracle_submit`. Do not pass `modelFamily`, `effort`, or `autoSwitchToThinking`.
+- If unsure which preset fits the task, ask the user.
+
 Rules:
 - Always include an archive. Do not submit without context files.
 - By default, include the whole repository by passing `.`. Default archive exclusions apply automatically, including common bulky outputs and obvious credentials/private data like `.env` files, key material, credential dotfiles, local database files, and root `secrets/` directories.
@@ -21,8 +27,7 @@ Rules:
 - If the request depends on git state or pending changes (for example code review, ship readiness, or release approval), create a tracked diff bundle file inside the repo (for example under `.pi/`) containing `git status` plus `git diff` output, include that file in the archive, and tell the oracle to use it because the `.git` directory is not included in oracle exports.
 - When `files=["."]` and the post-exclusion archive is still too large, submit automatically prunes the largest nested directories matching generic generated-output names like `build/`, `dist/`, `out/`, `coverage/`, and `tmp/` outside obvious source roots like `src/` and `lib/` until the archive fits or no candidate remains. Successful submissions report what was pruned.
 - If a submitted oracle job later fails because upload is rejected, retry with a smaller archive in this order: (1) remove the largest obviously irrelevant/generated content, (2) if still too large, include modified files plus adjacent files plus directly relevant subtrees, (3) if still too large, explain the cut or ask the user.
-- Prefer the configured default model/effort unless the task clearly needs something else.
-- Only use autoSwitchToThinking with the instant model family.
+- Prefer the configured default (omit **`preset`**) unless the task clearly needs a different model; then choose a **`preset`** id from the tool schema enum.
 - If `oracle_submit` itself fails because the local archive still exceeds the upload limit after default exclusions and automatic generic generated-output-dir pruning, or for any other submit-time error, stop and report the error. Do not retry automatically.
 - If `oracle_submit` returns a queued job instead of an immediately dispatched one, treat that as success and end your turn exactly the same way.
 - After oracle_submit returns, end your turn. Do not keep working while the oracle runs.