@jameslovespancakes/pi-plus 1.0.0 → 1.0.1

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

@@ -1,86 +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.
+---
+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

@@ -1,10 +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>
+<?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

@@ -1,67 +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"
-  }
-}
+{
+  "name": "@jameslovespancakes/pi-plus",
+  "version": "1.0.1",
+  "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"
+  }
+}