@jameslovespancakes/pi-plus 1.0.0

package/LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2026 james
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

package/README.md

@@ -0,0 +1,190 @@
+<div align="center">
+
+<img src="images/pi-plus.svg" alt="pi-plus" width="140">
+
+# pi-plus
+
+**Everything [pi](https://pi.dev/) is missing, in one install.**
+
+[![npm](https://img.shields.io/npm/v/@jameslovespancakes/pi-plus?color=%234D9ABF&label=npm)](https://www.npmjs.com/package/@jameslovespancakes/pi-plus)
+[![license](https://img.shields.io/badge/license-MIT-F1BE58)](LICENSE)
+[![pi-package](https://img.shields.io/badge/pi--package-F09082)](https://pi.dev/packages)
+
+</div>
+
+---
+
+## Install
+
+```sh
+pi install npm:@jameslovespancakes/pi-plus           # npm
+pi install git:github.com/jameslovespancakes/pi-plus # git
+pi install /path/to/pi-plus                          # local checkout
+```
+
+Then run **`/pi-plus`**. It shows what is configured and what is not, and enter
+on any row starts that setup.
+
+<img src="images/pi-plus_demo.png" alt="The /pi-plus status modal listing each feature and its state" width="700">
+
+`/pi-plus help` asks the model to explain the pack and what you are missing.
+Nothing else is required, every feature configures itself from within pi.
+
+---
+
+## Features
+
+### Pool every subscription
+
+Add all the Claude accounts you own. pi-plus balances sessions across them by
+**remaining quota and time-to-reset**, keeps caches sticky, and migrates only on
+confirmed exhaustion.
+
+```
+/account anthropic add work
+/routing optimal
+```
+
+Your live quota, always in the footer:
+
+<img src="images/usage_demo.png" alt="Claude subscription quota bars in the pi footer" width="680">
+
+With more than two accounts only the two most recently used are listed, so the
+footer stays a fixed height however many you pool.
+
+Account and routing commands are provider-agnostic: they dispatch through an
+adapter registry, so a second provider is one adapter, not a new command
+surface. Anthropic ships today.
+
+### Pick models on evidence
+
+`list_models` puts the full **Artificial Analysis** benchmark set in front of the
+agent: intelligence, coding, Terminal-Bench, τ²-bench, GPQA, price, tok/s and
+score-per-dollar, *plus your remaining quota*.
+
+```
+model                         bill quota  intel   code tbHard  $/1M  tok/s code/$
+anthropic/claude-opus-5       sub   74%    71.2   76.4   42.1     -      -      -
+openai-codex/gpt-6-astra      sub   81%    69.8   74.1   39.6     -      -      -
+openrouter/glm-5              paid    -    64.3   68.9   31.2  0.62   88.4  111.1
+```
+
+Cached locally and revalidated every 4h with an ETag, mirroring how pi refreshes
+its own model catalog.
+
+### Never get surprise-billed
+
+Metered providers are gated **at the provider boundary**, not by prompt
+instructions, so it covers workflow subagents too. `/provider` shows every
+provider you have credentials for and toggles each one:
+
+<img src="images/provider_demo.png" alt="The /provider picker toggling provider approval" width="560">
+
+### Run tests on real hardware
+
+`/remote setup` reads every connectable host in `~/.ssh/config` and lets you
+toggle which are eligible. No SSH config? It generates a dedicated key, shows
+the one line to run, and verifies.
+
+<img src="images/remote_demo.png" alt="The /remote picker toggling SSH workers" width="700">
+
+`remote_test` snapshots your working tree, admission-checks CPU/GPU/disk,
+reserves a slot, and **deletes the uploaded source the moment the run ends**,
+keeping only `test.log` and `result.json`. Cleanup runs on the worker, so it
+still happens if your laptop sleeps.
+
+### Coordinate multiple agents
+
+`agent_board` gives live presence, messaging, repo rooms and coordinator chains
+across every running pi agent. `/board` opens the messaging view:
+
+<img src="images/board_demo.png" alt="The /board messaging view" width="700">
+
+`/board setup` installs the server locally (pi starts it each session) or onto
+any Mac or Linux host over SSH, where launchd or systemd brings it back after a
+reboot.
+
+---
+
+## Commands
+
+| | |
+| --- | --- |
+| `/pi-plus` | status modal for every feature |
+| `/pi-plus help` | model explains the pack and what is missing |
+| `/account` | account hub: toggle, add, reauth, switch routing |
+| `/account <provider> add [label]` | add a subscription |
+| `/account <provider> reauth [label]` | reauthorize one |
+| `/routing standard \| optimal` | main-first, or balance by quota |
+| `/usage [on\|off\|text]` | quota bars |
+| `/models [sort]` | ranked catalog |
+| `/model-info <id>` | every benchmark for one model |
+| `/model-info refresh` | force a benchmark refresh |
+| `/model-info setup` | add the Artificial Analysis key |
+| `/provider` | provider toggle picker |
+| `/provider approve \| remove <name>` | grant or revoke |
+| `/remote setup` | worker hub: toggle, add, rename, remove |
+| `/remote add \| rename \| remove` | jump to one step |
+| `/board` | live agent board UI |
+| `/board setup \| restart \| clear \| status` | manage the board server |
+
+**Tools available to the agent:** `list_models`, `agent_board`, `remote_status`,
+`remote_test`.
+
+---
+
+## Configuration
+
+Everything lives in one file, `~/.pi/agent/pi-plus.json`, created on first use:
+
+```json
+{
+  "env":    { "ARTIFICIAL_ANALYSIS_API_KEY": "aa_…", "AGENT_BOARD_URL": "ws://…" },
+  "policy": { "autoApprove": [], "requireApproval": [], "deny": [] },
+  "remote": { "workers": [] }
+}
+```
+
+Real environment variables of the same name always win, so CI and shell
+overrides work unchanged. Older `model-quality-key.json`, `agent-board.json` and
+`remote-workers.json` are migrated automatically and left in place.
+
+---
+
+## Architecture
+
+Layered by dependency direction, grouped by domain. Each domain owns its own
+commands, tools and widgets.
+
+```
+src/
+  core/       zero pi imports, pure and unit-testable
+              store · config · env · quota · catalog · policy · exec · accounts
+  services/   stateful singletons with lifecycle + subscribe()
+              usage-service, the single quota poller
+  ui/         dumb render primitives: format · usage-bars
+  domains/    one pi extension entry each
+              setup · subscriptions · models · agents · remote
+  vendor/     anthropic.ts, the only file importing @cortexkit/*
+server/       the agent board server, one file and one dependency (ws)
+```
+
+```sh
+npm install && npm run verify   # typecheck + 106 tests
+```
+
+---
+
+## Built on
+
+pi-plus is a thin layer over other people's work.
+
+| project | what it does here | license |
+| --- | --- | --- |
+| [`pi`](https://pi.dev/)| the host agent and the entire extension API | MIT |
+| [`@cortexkit/pi-anthropic-auth`](https://github.com/cortexkit/anthropic-auth) | Claude OAuth, the Anthropic provider, token refresh and routing primitives | MIT |
+
+
+## License
+
+MIT

package/config/pi-plus.example.json

@@ -0,0 +1,60 @@
+{
+  "env": {
+    "ARTIFICIAL_ANALYSIS_API_KEY": "<artificial-analysis-api-key>",
+    "AGENT_BOARD_URL": "ws://<board-host>:8787/ws",
+    "AGENT_BOARD_TOKEN": "<agent-board-token>",
+    "AGENT_BOARD_NAME": "you"
+  },
+  "policy": {
+    "autoApprove": [
+      "anthropic/*",
+      "openai-codex/*",
+      "openrouter/stealth/union-alpha"
+    ],
+    "requireApproval": [
+      "openrouter/*",
+      "google/*",
+      "openai/*",
+      "xai/*"
+    ],
+    "deny": []
+  },
+  "remote": {
+    "injectStatus": true,
+    "defaults": {
+      "cpuBlockPercent": 90,
+      "gpuBlockPercent": 90,
+      "gpuMemoryBlockPercent": 90,
+      "memoryBlockPercent": 90,
+      "minimumFreeDiskGB": 10,
+      "statusCacheSeconds": 30,
+      "retentionHours": 24
+    },
+    "workers": [
+      {
+        "name": "mac-mini",
+        "ssh": "mac-mini",
+        "root": "~/remote_tests",
+        "nice": 10,
+        "tags": [
+          "macos",
+          "arm64",
+          "gpu",
+          "metal"
+        ]
+      },
+      {
+        "name": "hyperion",
+        "ssh": "hyperion",
+        "root": "~/remote_tests",
+        "nice": 10,
+        "tags": [
+          "linux",
+          "x64",
+          "gpu",
+          "cuda"
+        ]
+      }
+    ]
+  }
+}

package/config/skills/model-routing/SKILL.md

@@ -0,0 +1,86 @@
+---
+name: model-routing
+description: Choose the right model for a task or workflow stage using live benchmark data. Use when delegating work to subagents, authoring or running inline workflows, or whenever deciding which model should handle a task.
+---
+
+# Model routing
+
+Do not guess model quality and do not rely on fixed `small`/`medium`/`big`
+profiles. Those profiles have been removed from this machine. Pick a model per
+task from live data.
+
+## 1. Read the catalogue first
+
+Call the `list_models` tool before delegating work:
+
+```
+list_models(sort_by: "coding", limit: 15)
+list_models(sort_by: "cost_efficiency", provider: "openai-codex")
+list_models(query: "opus", detail: "full")
+```
+
+It returns, per model:
+
+- Artificial Analysis Intelligence, Coding and Math indices
+- Terminal-Bench Hard, Terminal-Bench 2.1, τ²-bench, τ-bench Banking
+- LiveCodeBench, SciCode, GPQA Diamond, HLE, MMLU-Pro, IFBench, AIME, MATH-500,
+  long-context reasoning
+- price per 1M input/output/blended tokens, throughput, time to first token
+- cost-efficiency ratios (score per dollar)
+- `billing` (`sub` or `paid`) and remaining subscription quota
+- `conf`: `measured`, `inferred`, or `unrated`
+
+## 2. Match the metric to the task
+
+| Task | Rank by |
+|---|---|
+| Agentic coding, terminal work, multi-step edits | Terminal-Bench Hard, then Coding Index |
+| Tool-calling reliability | τ²-bench |
+| Algorithmic code generation | LiveCodeBench, SciCode |
+| Research, analysis, hard reasoning | GPQA, HLE, Intelligence Index |
+| Long document or repo-wide reasoning | long-context reasoning, context window |
+| Bulk or throwaway fan-out | cost efficiency and tok/s |
+
+## 3. Respect billing policy
+
+- Subscription models (`anthropic/*`, `openai-codex/*`) are free to use. Prefer
+  them, and prefer accounts with quota remaining.
+- Metered models (`openrouter/*`, `google/*`, `openai/*`, `xai/*`) are blocked at
+  the provider boundary until the user runs `/model-approve <provider>`. If a
+  call is blocked, switch to a subscription model or ask the user; never retry
+  in a loop.
+- Check `/model-policy` for the current spend ceiling.
+
+## 4. Use explicit models in workflows
+
+In inline workflow scripts pass `model` on each `agent()` call:
+
+```ts
+const findings = await parallel([
+  () => agent(prompt, { model: "openai-codex/gpt-5.6-terra", schema: Finding }),
+  () => agent(prompt, { model: "anthropic/claude-sonnet-5", schema: Finding }),
+]);
+
+const synthesis = await agent(summaryPrompt, {
+  model: "anthropic/claude-opus-5",
+  thinkingLevel: "high",
+});
+```
+
+Rules:
+
+- Always use the full `provider/id` form.
+- Give cheap, fast, high-throughput models the wide fan-out stages.
+- Give the strongest agentic model the synthesis, planning and patch-writing
+  stages.
+- Thinking level changes measured quality: `list_models` scores resolve per
+  effort level where Artificial Analysis publishes them, so raise
+  `thinkingLevel` when a stage is hard rather than switching to a costlier model
+  by default.
+- If a model is `unrated`, say so before using it for important work, and prefer
+  a `measured` alternative of similar cost.
+
+## 5. Quota awareness
+
+If Claude's 5h pool is nearly exhausted, route to Codex instead of stalling, and
+mention the switch. The `/usage` bar shows remaining quota for every account.

package/images/pi-plus.svg

@@ -0,0 +1,10 @@
+<?xml version="1.0" encoding="UTF-8"?>
+<svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 800 800">
+  <!-- Original Pi logo paths, unchanged -->
+  <path fill="#F09082" d="M165.29 165.29H517.36V400H400V282.65H165.29Z"/>
+  <path fill="#4D9ABF" d="M165.29 282.65H282.65V400H400V517.36H282.65V634.72H165.29Z"/>
+  <path fill="#F1BE58" d="M517.36 400H634.72V634.72H517.36Z"/>
+
+  <!-- Plus mark, matching the logo's square/pixel geometry -->
+  <path fill="#F09082" d="M697.84 223.97H736.96V263.09H776.08V302.21H736.96V341.33H697.84V302.21H658.72V263.09H697.84Z"/>
+</svg>

package/package.json

@@ -0,0 +1,67 @@
+{
+  "name": "@jameslovespancakes/pi-plus",
+  "version": "1.0.0",
+  "type": "module",
+  "description": "Consolidated pi extensions: Claude subscription accounts and quota HUD, benchmark-driven model catalog and billing policy, live agent board, and capacity-gated remote test workers.",
+  "license": "MIT",
+  "author": "james",
+  "keywords": [
+    "pi-package",
+    "pi",
+    "extensions",
+    "anthropic",
+    "claude",
+    "quota",
+    "models"
+  ],
+  "repository": {
+    "type": "git",
+    "url": "git+https://github.com/jameslovespancakes/pi-plus.git"
+  },
+  "homepage": "https://github.com/jameslovespancakes/pi-plus#readme",
+  "bugs": {
+    "url": "https://github.com/jameslovespancakes/pi-plus/issues"
+  },
+  "files": [
+    "src",
+    "server",
+    "config/skills",
+    "config/pi-plus.example.json",
+    "images",
+    "README.md",
+    "LICENSE"
+  ],
+  "pi": {
+    "extensions": [
+      "./src/domains/setup/index.ts",
+      "./src/domains/subscriptions/index.ts",
+      "./src/domains/models/index.ts",
+      "./src/domains/agents/index.ts",
+      "./src/domains/remote/index.ts"
+    ],
+    "skills": [
+      "./config/skills"
+    ]
+  },
+  "dependencies": {
+    "ws": "^8.18.3"
+  },
+  "peerDependencies": {
+    "@earendil-works/pi-agent-core": "*",
+    "@earendil-works/pi-ai": "*",
+    "@earendil-works/pi-coding-agent": "*",
+    "@earendil-works/pi-tui": "*",
+    "typebox": "*"
+  },
+  "scripts": {
+    "lint": "oxlint",
+    "lint:fix": "oxlint --fix",
+    "typecheck": "tsc -p tsconfig.json",
+    "test": "node --test \"tests/*.test.ts\"",
+    "verify": "npm run lint && npm run typecheck && npm test",
+    "version:next": "node scripts/bump-version.mjs --dry-run"
+  },
+  "devDependencies": {
+    "oxlint": "^1.83.0"
+  }
+}