@99percentpeople/pi-codex-api 0.1.3 → 0.2.0

package/README.md

@@ -1,216 +1,130 @@
 # @99percentpeople/pi-codex-api
 
-Expose first-party Codex subscription APIs as native Pi tools and commands.
-The extension reuses Pi's existing `openai-codex` OAuth login and does not need
-an OpenAI API key, MCP server, or separate search provider.
-
-## Requirements
-
-- Pi authenticated with `/login` for the `openai-codex` provider
-- An active `openai-codex` model, or **Other providers** enabled in `/99settings`
-- A ChatGPT workspace and plan entitled to the requested Codex feature
-
-The extension still loads when Codex has not been configured. It does not
-prompt or attempt an OAuth request during installation. A tool call instead
-shows an actionable error directing you to `/login`; an expired login is
-reported the same way after Pi's refresh attempt fails.
-
-API-key authentication is intentionally rejected because these tools target the
-ChatGPT subscription backend rather than metered Platform API usage.
-
-## Install
+Turn your ChatGPT subscription into Pi superpowers — image generation, web
+search, Fast mode, and usage monitoring — **no OpenAI API key required**.
+
+## Highlights
+
+- **Image generation & editing** — `codex_image` creates or edits images via
+  your Codex subscription's `gpt-image-2`, and works from **any model**:
+  even with a third-party provider active (DeepSeek, Google, …), it reuses Pi's
+  logged-in `openai-codex` OAuth account (enable **Other providers** in
+  `/99settings`).
+- **First-party search** — `codex_search` runs web and image queries, page
+  navigation, PDF screenshots, finance, weather, sports, and time lookups, and
+  renders clean result cards instead of raw citations.
+- **Fast mode** — optional priority service tier for snappier responses.
+- **Usage at a glance** — the status bar shows remaining Codex quota and
+  reset time; `/codex-usage` shows your plan, masked account, limits, credits,
+  and earned rate-limit reset cards.
+- **Reset cards** — when you run out of messages, `/codex-redeem` redeems an
+  earned reset card safely: pick a card, confirm, done.
+
+## Quick start
 
 ```bash
 pi install npm:@99percentpeople/pi-codex-api
 ```
 
-During development, build the npm artifact before installing the package
-directory:
+Requirements:
 
-```bash
-bun run build:packages
-bun run --cwd extensions/codex-api build
-pi install ./extensions/codex-api
-```
+- Pi logged in with `/login` for `openai-codex`
+- An active `openai-codex` model, or **Other providers** enabled in
+  `/99settings` to use the subscription from any model
 
-## Tools
+That's it — just ask *"generate an image of a neon ramen shop"* or *"search
+the web for today's Rust releases"* and the model calls the tools for you.
+
+## Commands
 
-### `codex_image`
+### `/codex-usage`
 
-Generates or edits images through the Codex subscription image API using
-`gpt-image-2`.
+Shows a fresh snapshot of your subscription in one block:
 
 ```text
-codex_image prompt="A quiet neon-lit ramen shop at night"
-```
+Codex usage
 
-For edits, provide up to five image paths inside the current workspace:
+account · Plus (user@example.com)
 
-```text
-codex_image \
-  prompt="Change only the sky to a warm sunset" \
-  referenced_image_paths=["assets/photo.png"] \
-  output_path="assets/photo-sunset.png"
+codex
+  weekly [█████████████░░░░░░░] 65% left resets in 5d 3h
+  no additional credits
+
+rate limit redeem
+  Full reset (available, expires 2026-08-13 02:14 UTC+8)
 ```
 
-If an edit target was attached or generated in the conversation and has no
-usable local path, the model can instead include the smallest necessary window
-of one to five recent conversation images with `num_last_images_to_include`.
-The two reference mechanisms are mutually exclusive.
-
-The model may request a validated GPT Image 2 `WIDTHxHEIGHT` size when exact
-dimensions are part of the task. It normally omits `quality`, inheriting the
-user's **Image quality** preference, and only overrides it when the user
-explicitly requests a draft or a quality level.
-
-New images default to `output/codex-images/<tool-call-id>.png`. Reference and
-output paths must stay inside the current workspace, and existing files are
-never overwritten. The PNG is both saved locally and returned as image
-content so Codex can inspect or revise it in later turns. If a network request
-fails before an HTTP response, the error identifies the endpoint path and any
-safe transport code (for example `ECONNRESET`). The extension itself never
-retries image generation automatically, because the server may have accepted
-the first request; decide whether to retry after reviewing the error.
-
-### `codex_search`
-
-Uses the first-party Codex standalone search API. Supported command families:
-
-- web and image queries;
-- open, click, and find operations using returned reference IDs;
-- PDF page screenshots;
-- finance, weather, sports, and time lookups.
-
-Search mode is configurable as Auto, Cached, Indexed, or Live. In Auto, the AI
-requests a per-call mode: Cached for stable facts and known references, Indexed
-for recent documentation and announcements, and Live for same-day or real-time
-information. An omitted request defaults to Indexed. Selecting a fixed mode
-pins every call to that mode, regardless of the AI request. External content is
-untrusted and should never be treated as instructions.
-
-## Parameter ownership
-
-The extension follows the current official Codex split between task intent,
-user policy, and internal protocol fields:
-
-| Owner | Search | Image |
-| --- | --- | --- |
-| AI per call | queries, recency, task-specific domains, navigation/lookup commands, response length, and the requested mode when Auto is enabled | prompt, local or recent-conversation references, task-specific size, explicit quality override, destination path |
-| User in `/99settings` | Auto or a fixed Cached/Indexed/Live policy, plus search context size | default image quality |
-| Extension/backend | session/model routing, caller policy and token ceiling | fixed `gpt-image-2`, automatic background, one PNG result, reference and dimension validation |
-
-Model selection, batch count, output format, input fidelity, masks, moderation,
-and transparent background are intentionally not advertised as AI arguments on
-this subscription path. In particular, GPT Image 2 does not support native
-transparent-background output; requesting transparency through an unsupported
-field would be misleading.
-
-## Bundled skill
-
-The package also ships the Agent Skills-standard **`gpt-image-prompts`** skill
-under `skills/`. Pi discovers it with the package and can load it on demand to
-write or refine production-ready GPT Image 2 prompts for new images,
-reference-guided work, and precise edits. Invoke it explicitly with:
+- Plan type and masked email from Codex's official usage endpoint
+- Each limit window shows remaining capacity as a bar; when exhausted it reads
+  `limit reached` instead of `0% left`
+- Earned reset cards, sorted by expiry, with local-time expiry timestamps
+- The status bar mirrors this compactly: `Codex weekly 65% 5d 3h` →
+  `Codex weekly limit reached 5d 14h`
 
-```text
-/skill:gpt-image-prompts
-```
+Usage refreshes automatically on session start and model select;
+`/codex-usage` always forces a fresh read.
 
-The skill contains visual prompt-writing guidance only. It does not select or
-document tools, construct API requests, manage credentials, or prescribe file
-and execution workflows.
-
-## Tool display
-
-Web and image queries render normalized source cards instead of the backend's
-raw citation dump. The collapsed view shows three title/domain/snippet cards;
-Pi's configured tool-output expansion shortcut (`Ctrl+O` by default) expands
-every source with its full URL. Internal reference markers,
-word-limit metadata, and separators are hidden. Open/click/find/PDF operations
-use cleaned document cards. Batched navigation renders each returned page as a
-separate numbered card with a short per-page preview, one shared expansion hint,
-and visible warning styling for unresolved references. Weather, finance,
-sports, and time lookups use a compact data view. This changes only the TUI display copy: the model
-still receives the complete original search output.
-
-Search and image parameters stream into the call row while the model constructs
-them; input parameters are never repeated in the result area. Both tools also
-stream real execution stages into the active result row while they run
-(authentication, request, reference loading, generation, and saving as
-applicable). Successful `codex_search` and `codex_image` calls also request a
-rate-limited background usage refresh. The Codex search and image endpoints
-return one final response, so result bodies appear atomically rather than as
-fabricated content chunks.
+### `/codex-redeem`
 
-## Commands
+Redeems an earned rate-limit reset card when you're out of messages:
 
 ```text
-/codex-usage
+──────────────────────────────────────────────
+Select a reset credit to redeem (30s)
+
+→ Full reset (expires 2026-08-03)
+  Full reset (expires 2026-08-12)
+
+↑↓ navigate  ↵ select  esc cancel
+──────────────────────────────────────────────
+      ↓ pick a card
+──────────────────────────────────────────────
+Redeem Full reset (expires 2026-08-12)? (30s)
+
+→ No
+  Yes
+
+↑↓ navigate  ↵ select  esc cancel
+──────────────────────────────────────────────
 ```
 
-Fast mode is controlled only through **Codex API → Fast mode** in
-`/99settings`. It sends `service_tier: "priority"` on Codex Responses requests
-and can reduce latency while consuming included limits faster. The extension fetches
-subscription usage directly from Codex's official ChatGPT WHAM endpoint when a
-session starts or a Codex model is selected; `/codex-usage` forces a fresh
-read, so no model request is required first. Automatic post-response refreshes
-are limited to once per minute.
-
-Pi does not currently publish an OAuth-account-change event or expose its
-credential store to extensions. As a temporary compatibility layer, this
-extension watches Pi's agent-directory `auth.json`. A short debounce reloads
-the public model registry and compares the resolved Codex Account ID. After
-`/login` replaces the account, the old snapshot is removed, the status changes
-to `Codex syncing…`, and a forced account-scoped refresh starts. `/logout`
-clears the status. In-flight requests are revision-guarded, so an old account
-cannot overwrite the newly active account. The watcher is closed on session
-teardown and never accesses Pi's private authentication runtime.
-
-`/codex-usage` groups every metered limit under a simple `Codex usage` heading.
-Each server-provided window (for example 5h, daily, or weekly) uses a fixed
-20-cell bar whose filled portion represents remaining capacity, followed only
-by `% left` and its reset time. Additional credit availability stays under the
-same limit group. Inactive zero-value placeholder windows are hidden. Response
-headers remain supported as a fallback.
+- Multiple cards show a picker (earliest expiry first); a single card skips to
+  confirmation
+- **No is the default**, so a stray Enter never consumes anything
+- Redemptions are idempotent: a retry after a network failure can never consume
+  a second card
+- Without dialog UI, it falls back to a two-step confirm flow
 
 ## Settings
 
-Use the shared menu:
+Configure under **Codex API** in `/99settings`:
 
-```text
-/99settings
-```
+- **Other providers** — let any model (DeepSeek, Google, …) use the logged-in
+  Codex subscription
+- **Fast mode** — priority service tier (lower latency, faster limit use)
+- **Search** — auto routing (Cached / Indexed / Live per call) or a fixed mode,
+  plus context size
+- **Image quality** — default GPT Image 2 quality
+- **Usage status** — show or hide the Codex quota line in the status bar
 
-The **Codex API** section controls:
+Settings live in `~/.pi/agent/99extensions.json` under the `codex-api`
+namespace.
 
-- **Other providers** — off by default. When enabled, a non-Codex model may
-  call `codex_image` and `codex_search`; the extension resolves only the
-  separately logged-in `openai-codex` OAuth account. It never sends the active
-  model provider's credentials to ChatGPT.
-- **Fast mode** — enables or disables the priority service tier;
-- Auto search routing, or a fixed Cached, Indexed, or Live policy;
-- search context size;
-- default GPT Image 2 quality (`Auto`, `Low`, `Medium`, or `High`);
-- subscription usage status visibility.
+## How it works
 
-Configuration is stored under the `codex-api` namespace in:
+- Reuses Pi's existing `openai-codex` OAuth — no API key, no MCP server, no
+  separate search provider
+- Tokens are fetched per call from Pi's model registry, never stored in
+  settings or tool results
+- Images and search requests go to OpenAI's Codex backend and follow your
+  ChatGPT workspace's policies
+- The extension also ships the **`gpt-image-prompts`** skill for crafting
+  production-grade image prompts — invoke it with `/skill:gpt-image-prompts`
 
-```text
-~/.pi/agent/99extensions.json
-```
+## Development
 
-## Authentication and privacy
-
-The extension asks Pi's model registry for a refreshed OAuth token at tool-call
-time. It sends the same bearer token and `ChatGPT-Account-ID` used by Pi's
-`openai-codex` adapter. With **Other providers** enabled, it resolves that
-Codex model specifically rather than using the active model's credentials.
-Tokens are never copied into extension settings or tool results. The extracted
-ChatGPT account ID is used only as an in-memory Usage-state key and is never
-persisted or displayed. The watcher reacts only to the `auth.json` filename;
-credential parsing and resolution remain inside Pi's public model registry.
-
-Image prompts, reference images, search commands, and search context are sent
-to OpenAI's Codex backend and are subject to the active ChatGPT workspace's
-policies. Feature availability and request formats may change as Codex rolls
-out backend updates.
+```bash
+bun run build:packages
+bun run --cwd extensions/codex-api build
+pi install ./extensions/codex-api
+```