@diegopetrucci/pi-extensions 0.1.74 → 0.1.78

package/extensions/dynamic-context-pruning/package.json

@@ -1,6 +1,6 @@
 {
   "name": "pi-dynamic-context-pruning",
-  "version": "0.1.11",
+  "version": "0.1.13",
   "type": "module",
   "description": "A pi extension that dynamically prunes stale context to keep sessions lean.",
   "keywords": [

package/extensions/fast/README.md

@@ -1,6 +1,6 @@
 # fast
 
-A pi extension that provides one `/fast` toggle for supported direct OpenAI API, OpenAI Codex, and Anthropic Claude models. It selects the provider-specific Fast-mode request shape from the currently selected model.
+A pi extension that provides `/fast` and `/ultrafast` toggles for supported direct OpenAI API, OpenAI Codex, and Anthropic Claude models. It selects the provider-specific Fast-mode request shape from the currently selected model.
 
 For eligible direct OpenAI API models it injects:
 
@@ -44,12 +44,35 @@ Direct OpenAI API Fast mode requires all of the following:
 
 The direct OpenAI route stays on OpenAI's direct API contract for both API-key
 auth and Pi's `openai` ChatGPT OAuth. Acceptance of `service_tier: "fast"` by
-Pi's `openai` ChatGPT OAuth sign-in endpoint has not been live-verified and
-remains deferred to manual gate `pel-71t6`. This direct route is never forwarded
-to legacy `openai-codex` or its subscription-usage endpoint. The documented
-request shape is support evidence, not live certification; this change was
-validated only with mocked eligibility and request payloads, not live or paid
-provider calls.
+Pi's `openai` ChatGPT OAuth sign-in endpoint has not been live-verified.
+This direct route is never forwarded to legacy `openai-codex` or its
+subscription-usage endpoint. The documented request shape is support evidence,
+not live certification; this change was validated only with mocked eligibility
+and request payloads, not live or paid provider calls.
+
+### Ultrafast support (initial, fail-closed)
+
+Ultrafast is intentionally narrower than ordinary Fast mode. It requires all of
+the following on every activation and provider request:
+
+- Provider `openai`.
+- API `openai-responses`.
+- Model `gpt-6-astra`.
+- Direct API-key auth, not Codex OAuth or direct OpenAI OAuth.
+- Base URL exactly `https://api.openai.com/v1` (a single trailing slash is also accepted).
+
+It does not support Chat Completions, preview models, regional or proxy endpoints,
+or Anthropic Ultrafast. Eligible payloads receive the requested
+`service_tier: "ultrafast"` only when they do not already contain a
+`service_tier` field. Existing values and mismatched/malformed payloads are left
+untouched. The extension reports a requested tier; it does not claim that OpenAI
+accepted or applied it, and it never silently falls back to ordinary Fast.
+
+OpenAI documents this request shape in the [Ultrafast API guide](https://developers.openai.com/api/docs/guides/ultrafast-mode)
+and lists the account/model pricing in its [API pricing](https://developers.openai.com/api/docs/pricing).
+Actual Ultrafast API pricing is **6x Standard**, while Pi's current cost display
+omits that premium. Treat the activation warning as a deliberate paid-usage
+warning.
 
 OpenAI Codex Fast mode requires all of the following:
 
@@ -91,11 +114,26 @@ Anthropic Fast mode supports both API-key and Claude Code OAuth access when the
 
 ```text
 /fast
+/ultrafast
+/fast ultrafast
 ```
 
-Run `/fast` to toggle Fast mode for the current session/runtime. The enabled state follows model changes: switching between eligible OpenAI and Anthropic models automatically changes the request format without resetting the toggle. The footer shows `fast` only while the selected model is eligible and Fast mode is enabled.
+Run `/fast` to toggle ordinary Fast mode for the current session/runtime. The
+enabled state follows model changes: switching between eligible OpenAI and
+Anthropic models automatically changes the request format without resetting the
+toggle. The footer shows `fast` only while the selected model is eligible and
+Fast mode is enabled.
+
+`/ultrafast` and `/fast ultrafast` are equivalent toggles for the session-only
+Ultrafast override. Ultrafast defaults off, is never enabled by global or project
+configuration, and can be undone with either command. Running ordinary `/fast`
+while Ultrafast is selected turns Ultrafast off rather than silently switching to
+another paid tier. Switching to an unsupported model or endpoint disables
+Ultrafast and does not re-enable it when `gpt-6-astra` is selected again. The
+footer uses `ultrafast` while an eligible request is selected; status text says
+that the tier is requested, not remotely applied.
 
-The extension defaults to off so installing the full collection does not accidentally spend Fast-mode credits.
+The extension defaults to off so installing the full collection does not accidentally spend Fast-mode credits. Ultrafast verification is mocked-only: tests use local eligibility checks, fake payloads, and a fake fetch through Pi's `openai-responses` `streamSimple`; they make no live or paid provider requests.
 
 ## Config
 
@@ -105,13 +143,13 @@ Optional global config:
 ~/<pi-config-dir>/agent/extensions/fast.json
 ```
 
-Optional project config:
+Optional project config, from the nearest ancestor of the working directory (the working directory itself counts):
 
 ```text
 <project>/<pi-config-dir>/fast.json
 ```
 
-Here `<pi-config-dir>` is Pi's runtime config directory name (`CONFIG_DIR_NAME`; `.pi` by default). Project config overrides global config after Pi reports that the project is trusted.
+Here `<pi-config-dir>` is Pi's runtime config directory name (`CONFIG_DIR_NAME`; `.pi` by default). The loader uses that nearest ancestor `fast.json`, which overrides global config after Pi reports that the project is trusted.
 
 ```json
 {
@@ -123,7 +161,7 @@ Here `<pi-config-dir>` is Pi's runtime config directory name (`CONFIG_DIR_NAME`;
 - `enabled`: default Fast-mode state when there is no session override.
 - `showStatus`: show a compact `fast` status when Fast mode is active for the selected model.
 
-The unified extension intentionally does not read `openai-fast.json` or `claude-fast.json`; those files belong to the provider-specific standalone packages and may contain conflicting defaults.
+The unified extension reads only `fast.json`. It does not read `openai-fast.json` or `claude-fast.json`.
 
 ## Install
 
@@ -153,7 +191,7 @@ Then reload pi:
 
 ## Legacy provider-specific packages
 
-The collection package loads this unified extension instead of `openai-fast` and `claude-fast`. Both provider-specific packages remain available as standalone alternatives. Do not load them together with this extension: independent Fast-mode state and duplicate `/fast` registrations can produce conflicting behavior.
+The collection package loads this unified extension only (`extensions/fast/index.ts`). `openai-fast` and `claude-fast` remain published as deprecated standalone leftovers. On session start they warn, clear their leftover footer status, and remove `fast-mode-2026-02-01` from the current model header. They do not read config, register commands, or inject Fast fields into requests.
 
 ## Notes
 
@@ -163,6 +201,6 @@ The collection package loads this unified extension instead of `openai-fast` and
 - Codex Fast availability depends on plan, client, workspace settings, and rollout. Local eligibility is not live-provider certification.
 - Pi's provider catalog remains the source of reported model pricing, including any tiered pricing; this extension does not apply guessed discounts or rewrite cost metadata.
 - OpenAI Codex Fast mode intentionally does not affect API-key models.
-- The retired `openai-fast` and `claude-fast` packages remain legacy standalone boundaries; new provider/auth behavior belongs only to this unified extension.
+- Deprecated `openai-fast` and `claude-fast` packages no longer implement Fast mode. New provider/auth behavior belongs only to this unified extension.
 - Existing `speed` and `service_tier` fields are never overwritten.
 - Pi's model catalog may not include a Fast-mode premium. Cost accounting depends on the provider reporting the effective tier in its streamed response, and this extension does not patch usage totals.