@opengsd/gsd-pi 1.0.1 → 1.0.2-dev.0ee79e5

package/src/resources/extensions/ollama/tests/ollama-chat-provider-think.test.ts

@@ -0,0 +1,87 @@
+// gsd-pi — Tests for ollama think-parameter mapping
+import { describe, it } from "node:test";
+import assert from "node:assert/strict";
+import type { Api, Context, Model, SimpleStreamOptions } from "@gsd/pi-ai";
+import { buildRequest, buildThinkParam } from "../ollama-chat-provider.js";
+
+function modelStub(reasoning: boolean, id = "gpt-oss:20b"): Model<Api> {
+	return {
+		id,
+		name: id,
+		api: "openai-completions" as Api,
+		provider: "ollama",
+		baseUrl: "http://localhost:11434",
+		reasoning,
+		input: ["text"],
+		cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 },
+		contextWindow: 131072,
+		maxTokens: 32768,
+	};
+}
+
+describe("buildThinkParam — gating", () => {
+	it("returns undefined when model.reasoning is false", () => {
+		const m = modelStub(false);
+		const opts: SimpleStreamOptions = { reasoning: "high" };
+		assert.equal(buildThinkParam(m, opts), undefined);
+	});
+
+	it("returns undefined when options.reasoning is undefined (preserve existing default)", () => {
+		const m = modelStub(true);
+		assert.equal(buildThinkParam(m, {}), undefined);
+		assert.equal(buildThinkParam(m, undefined), undefined);
+	});
+});
+
+describe("buildThinkParam — ThinkingLevel mapping", () => {
+	const m = modelStub(true);
+
+	it("maps 'minimal' to false (turn thinking off)", () => {
+		assert.equal(buildThinkParam(m, { reasoning: "minimal" }), false);
+	});
+
+	it("passes 'low' through as string", () => {
+		assert.equal(buildThinkParam(m, { reasoning: "low" }), "low");
+	});
+
+	it("passes 'medium' through as string", () => {
+		assert.equal(buildThinkParam(m, { reasoning: "medium" }), "medium");
+	});
+
+	it("passes 'high' through as string", () => {
+		assert.equal(buildThinkParam(m, { reasoning: "high" }), "high");
+	});
+
+	it("collapses 'xhigh' to 'high' (ollama caps at high)", () => {
+		assert.equal(buildThinkParam(m, { reasoning: "xhigh" }), "high");
+	});
+});
+
+describe("buildRequest — wire-level think field", () => {
+	const emptyContext: Context = { messages: [] } as unknown as Context;
+
+	it("sets request.think = false when reasoning='minimal' on a reasoning model", () => {
+		// Wire-level coverage: buildThinkParam returns false for 'minimal',
+		// and the buildRequest guard must let falsy-but-not-undefined values
+		// through. A naive `if (think) request.think = think` would drop
+		// 'minimal' entirely and leave the model's default thinking on.
+		const req = buildRequest(modelStub(true), emptyContext, { reasoning: "minimal" });
+		assert.equal(req.think, false);
+		assert.ok('think' in req, "'think' key must be present in request when minimal is set — guards against future `delete request.think` refactors");
+	});
+
+	it("sets request.think = 'high' when reasoning='high'", () => {
+		const req = buildRequest(modelStub(true), emptyContext, { reasoning: "high" });
+		assert.equal(req.think, "high");
+	});
+
+	it("omits request.think when reasoning is unset (preserve model default)", () => {
+		const req = buildRequest(modelStub(true), emptyContext, {});
+		assert.equal(req.think, undefined);
+	});
+
+	it("omits request.think on non-reasoning models even when reasoning is requested", () => {
+		const req = buildRequest(modelStub(false), emptyContext, { reasoning: "high" });
+		assert.equal(req.think, undefined);
+	});
+});

package/src/resources/extensions/ollama/tests/ollama-client-timeout-env.test.ts

@@ -1,4 +1,4 @@
-// GSD2 — Tests for OLLAMA_PROBE_TIMEOUT_MS / OLLAMA_REQUEST_TIMEOUT_MS env vars (#5003 / #4982)
+// gsd-pi — Tests for OLLAMA_PROBE_TIMEOUT_MS / OLLAMA_REQUEST_TIMEOUT_MS env vars (#5003 / #4982)
 //
 // Pinned defaults: 1500 ms probe, 10 000 ms request. The defaults must be
 // preserved when the env var is unset, empty, non-numeric, zero, or negative

package/src/resources/extensions/ollama/tests/ollama-client.test.ts

@@ -1,4 +1,4 @@
-// GSD2 — Tests for Ollama HTTP client
+// gsd-pi — Tests for Ollama HTTP client
 import { describe, it, beforeEach, afterEach } from "node:test";
 import assert from "node:assert/strict";
 import { getOllamaHost } from "../ollama-client.js";

package/src/resources/extensions/ollama/tests/ollama-discovery-priority.test.ts

@@ -0,0 +1,89 @@
+// gsd-pi — Tests for ollama-discovery /api/show priority and num_ctx sync
+import { describe, it } from "node:test";
+import assert from "node:assert/strict";
+import { discoverModels } from "../ollama-discovery.js";
+import type { OllamaModelInfo, OllamaShowResponse, OllamaTagsResponse } from "../types.js";
+
+function makeDeps(showResp: Partial<OllamaShowResponse>, modelInfo: Partial<OllamaModelInfo> = {}) {
+	return {
+		listModels: async (): Promise<OllamaTagsResponse> => ({
+			models: [{
+				name: "test-model:latest",
+				model: "test-model:latest",
+				modified_at: "",
+				size: 1_000_000,
+				digest: "abc",
+				details: { parent_model: "", format: "", family: "", families: [], parameter_size: "7B", quantization_level: "" },
+				...modelInfo,
+			} as OllamaModelInfo],
+		}),
+		showModel: async () => ({
+			modelfile: "",
+			parameters: "",
+			template: "",
+			details: { parent_model: "", format: "", family: "", families: [], parameter_size: "7B", quantization_level: "" },
+			model_info: {},
+			capabilities: [],
+			...showResp,
+		} as OllamaShowResponse),
+	};
+}
+
+describe("enrichModel — /api/show context priority", () => {
+	it("uses /api/show context_length over a stale KNOWN_MODELS value", async () => {
+		// llama3.1 in KNOWN_MODELS = 131072. If /api/show says 262144, trust it.
+		const deps = makeDeps({ model_info: { "llama.context_length": 262144 } }, { name: "llama3.1:8b" });
+		const [m] = await discoverModels(deps);
+		assert.equal(m.contextWindow, 262144);
+	});
+
+	it("falls back to KNOWN_MODELS when /api/show provides no context_length", async () => {
+		const deps = makeDeps({ model_info: {} }, { name: "llama3.1:8b" });
+		const [m] = await discoverModels(deps);
+		assert.equal(m.contextWindow, 131072); // KNOWN_MODELS llama3.1
+	});
+});
+
+describe("enrichModel — num_ctx sync with /api/show", () => {
+	it("syncs ollamaOptions.num_ctx with showContextWindow when /api/show wins", async () => {
+		const deps = makeDeps({ model_info: { "llama.context_length": 262144 } }, { name: "llama3.1:8b" });
+		const [m] = await discoverModels(deps);
+		assert.equal(m.ollamaOptions?.num_ctx, 262144,
+			"num_ctx must mirror the authoritative contextWindow; sending stale num_ctx defeats the priority flip");
+	});
+
+	it("preserves sibling ollamaOptions fields when /api/show flips num_ctx", async () => {
+		// Drive enrichModel with a synthetic capabilities stub: model name matches a known
+		// table entry, but we mock the table indirectly by injecting deps that simulate
+		// what enrichModel would receive. Since enrichModel resolves caps internally via
+		// getModelCapabilities, the cleanest assertion is at the discoverModels output:
+		// the returned ollamaOptions must contain ALL fields from caps.ollamaOptions plus
+		// the synced num_ctx — confirmed by checking the num_ctx is overridden AND the
+		// returned ollamaOptions object reference is NOT equal to caps.ollamaOptions (it
+		// must be a fresh object from the spread). This catches a naive replacement
+		// `ollamaOptions = { num_ctx: showContextWindow }` that drops siblings.
+		const deps = makeDeps({ model_info: { "llama.context_length": 262144 } }, { name: "llama3.1:8b" });
+		const [m] = await discoverModels(deps);
+		// Sanity: num_ctx flipped to /api/show value
+		assert.equal(m.ollamaOptions?.num_ctx, 262144);
+		// Real coverage: the returned object must be a spread, not a literal {num_ctx}.
+		// We verify this structurally by checking that every key from the original
+		// caps.ollamaOptions (looked up directly from the source table) is present.
+		// llama3.1 table currently only has num_ctx — if/when sibling fields are added,
+		// this test will catch a regression where the spread is removed.
+		// For now we pin the spread invariant: ollamaOptions must be the fresh
+		// shallow-spread object, not a reference to caps.ollamaOptions.
+		const { getModelCapabilities } = await import("../model-capabilities.js");
+		const tableCaps = getModelCapabilities("llama3.1:8b");
+		const tableNumCtx = tableCaps.ollamaOptions?.num_ctx;
+		assert.notEqual(tableNumCtx, 262144, "test precondition: table num_ctx differs from /api/show value");
+		assert.notEqual(m.ollamaOptions, tableCaps.ollamaOptions,
+			"returned ollamaOptions must be a fresh spread object, not a reference to the table — otherwise a future direct replacement `{num_ctx}` would silently drop sibling fields");
+	});
+
+	it("preserves KNOWN_MODELS num_ctx when /api/show returns no context_length", async () => {
+		const deps = makeDeps({ model_info: {} }, { name: "llama3.1:8b" });
+		const [m] = await discoverModels(deps);
+		assert.equal(m.ollamaOptions?.num_ctx, 131072); // unchanged from table
+	});
+});

package/src/resources/extensions/ollama/tests/ollama-discovery.test.ts

@@ -1,4 +1,4 @@
-// GSD2 — Tests for Ollama model discovery and enrichment
+// gsd-pi — Tests for Ollama model discovery and enrichment
 import { describe, it } from "node:test";
 import assert from "node:assert/strict";
 import { discoverModels } from "../ollama-discovery.js";
@@ -18,15 +18,19 @@ function showStub(modelInfo: Record<string, unknown>): OllamaShowResponse {
 	return { modelfile: "", parameters: "", template: "", details: EMPTY_DETAILS, model_info: modelInfo };
 }
 
+function showWithCapabilities(modelInfo: Record<string, unknown>, capabilities: string[]): OllamaShowResponse {
+	return { ...showStub(modelInfo), capabilities };
+}
+
 describe("discoverModels — context window resolution", () => {
-	it("uses known table context window without calling /api/show", async () => {
-		let showCalled = false;
+	it("uses known table context window when /api/show returns no context_length", async () => {
+		// /api/show is now called unconditionally (for capabilities detection), but
+		// when it returns no context_length the table still wins for contextWindow.
 		const models = await discoverModels({
 			listModels: async () => tagsStub("llama3.2:latest", "3B"),
-			showModel: async () => { showCalled = true; throw new Error("should not be called"); },
+			showModel: async () => showStub({}),
 		});
 		assert.equal(models[0].contextWindow, 131072);
-		assert.equal(showCalled, false);
 	});
 
 	it("uses context_length from /api/show model_info for unknown model", async () => {
@@ -52,4 +56,38 @@ describe("discoverModels — context window resolution", () => {
 		});
 		assert.equal(models[0].contextWindow, 8192);
 	});
-});
+});
+
+describe("enrichModel — capability detection via /api/show.capabilities", () => {
+	it("treats showCapabilities['thinking'] as reasoning=true for an unknown model", async () => {
+		const deps = {
+			listModels: async () => ({ models: [modelStub("novel-reasoning-model:cloud")] }),
+			showModel: async () => showWithCapabilities({ "novel.context_length": 131072 }, ["thinking", "completion"]),
+		};
+		const [m] = await discoverModels(deps);
+		assert.equal(m.reasoning, true, "showCapabilities['thinking'] must set reasoning=true even when KNOWN_MODELS has no entry");
+	});
+
+	it("treats empty showCapabilities as reasoning=false even when KNOWN_MODELS has a reasoning entry — false (defined) wins over ?? fallthrough", async () => {
+		// llama3.1 in KNOWN_MODELS has no reasoning entry — pick a name where the table WOULD claim reasoning if asked.
+		// deepseek-r1 has `reasoning: true` in KNOWN_MODELS. If /api/show says capabilities = [] (no thinking),
+		// the ?? chain must NOT fall through to caps.reasoning. Empty array's includes('thinking') is false,
+		// which is a *defined* boolean, so ?? does not advance.
+		const deps = {
+			listModels: async () => ({ models: [modelStub("deepseek-r1:7b", "7B")] }),
+			showModel: async () => showWithCapabilities({ "deepseek.context_length": 131072 }, []),
+		};
+		const [m] = await discoverModels(deps);
+		assert.equal(m.reasoning, false,
+			"showCapabilities=[] returns false from includes(), which is *defined*; ?? must stop there, NOT fall through to caps.reasoning=true");
+	});
+
+	it("treats showCapabilities['vision'] as input including 'image'", async () => {
+		const deps = {
+			listModels: async () => ({ models: [modelStub("novel-vision-model:cloud")] }),
+			showModel: async () => showWithCapabilities({ "vision.context_length": 8192 }, ["vision", "completion"]),
+		};
+		const [m] = await discoverModels(deps);
+		assert.ok(m.input.includes("image"), "showCapabilities['vision'] must include 'image' in model.input");
+	});
+});

package/src/resources/extensions/ollama/thinking-parser.ts

@@ -1,4 +1,4 @@
-// GSD2 — Ollama Extension: Stateful <think> tag stream parser
+// gsd-pi — Ollama Extension: Stateful <think> tag stream parser
 
 /**
  * Extracts <think>...</think> thinking blocks from a streaming text response.

package/src/resources/extensions/ollama/types.ts

@@ -1,4 +1,4 @@
-// GSD2 — Ollama API response types
+// gsd-pi — Ollama API response types
 
 /**
  * Type definitions for the Ollama REST API.
@@ -37,6 +37,12 @@ export interface OllamaShowResponse {
 	template: string;
 	details: OllamaModelDetails;
 	model_info: Record<string, unknown>;
+	/**
+	 * Model capability flags exposed by ollama 0.4+. Common values:
+	 * "completion", "tools", "vision", "thinking", "embedding", "insert".
+	 * Older ollama versions and some cloud-routed models may omit this field.
+	 */
+	capabilities?: string[];
 }
 
 // ─── /api/ps ────────────────────────────────────────────────────────────────
@@ -97,6 +103,13 @@ export interface OllamaChatMessage {
 	tool_calls?: OllamaToolCall[];
 	/** Tool name — required for role: "tool" messages to correlate results with calls. */
 	name?: string;
+	/**
+	 * Reasoning trace returned by ollama 0.4+ when `think` is enabled on the
+	 * request. This is a separate channel from `content`; older models (and
+	 * some local models without native thinking support) emit thinking inline
+	 * as `<think>...</think>` tags in `content` instead.
+	 */
+	thinking?: string;
 }
 
 export interface OllamaToolCall {
@@ -136,6 +149,13 @@ export interface OllamaChatRequest {
 		num_gpu?: number;
 	};
 	keep_alive?: string;
+	/**
+	 * Controls thinking on reasoning models (ollama 0.4+).
+	 * - `true`/`false` toggles thinking on/off (universal form)
+	 * - `"low" | "medium" | "high"` sets thinking strength (gpt-oss style)
+	 * Omit to use the model's default behaviour.
+	 */
+	think?: boolean | "low" | "medium" | "high";
 }
 
 export interface OllamaChatResponse {

package/src/resources/extensions/shared/cmux-events.ts

@@ -1,4 +1,4 @@
-// GSD-2 — Shared cmux event channel contracts
+// gsd-pi — Shared cmux event channel contracts
 // Copyright (c) 2026 Jeremy McSpadden <jeremy@fluxlabs.net>
 
 /**

package/src/resources/extensions/shared/gsd-phase-state.ts

@@ -1,5 +1,5 @@
 /**
- * GSD2 Phase State — cross-extension coordination
+ * gsd-pi Phase State — cross-extension coordination
  * Copyright (c) 2026 Jeremy McSpadden <jeremy@fluxlabs.net>
  *
  * Lightweight module-level state that GSD auto-mode writes to and the

package/src/resources/extensions/shared/interview-ui.ts

@@ -1,4 +1,4 @@
-// GSD2 — Shared interview round UI widget
+// gsd-pi — Shared interview round UI widget
 /**
  * Shared interview round UI widget.
  *

package/src/resources/extensions/shared/tests/gsd-phase-state.test.ts

@@ -1,4 +1,4 @@
-// GSD2 Shared Phase State Coordination Tests
+// gsd-pi Shared Phase State Coordination Tests
 
 import { describe, it, beforeEach } from "node:test";
 import assert from "node:assert/strict";

package/src/resources/extensions/shared/tests/interview-notes-loop.test.ts

@@ -1,4 +1,4 @@
-// GSD2 — Regression test for interview-ui "None of the above" notes loop
+// gsd-pi — Regression test for interview-ui "None of the above" notes loop
 // Copyright (c) 2026 Jeremy McSpadden <jeremy@fluxlabs.net>
 
 /**

package/src/resources/extensions/shared/tests/next-action-ui-hasui.test.ts

@@ -1,4 +1,4 @@
-// GSD2 — Regression test for next-action-ui ctx.hasUI short-circuit (bare /gsd lockup)
+// gsd-pi — Regression test for next-action-ui ctx.hasUI short-circuit (bare /gsd lockup)
 
 /**
  * Regression test for the bare /gsd lockup investigated in

package/src/resources/extensions/subagent/launch.ts

@@ -1,4 +1,4 @@
-// GSD-2 + Subagent launch contract and child process safety helpers.
+// gsd-pi + Subagent launch contract and child process safety helpers.
 
 import * as fs from "node:fs";
 import * as path from "node:path";

package/src/resources/extensions/subagent/run-store.ts

@@ -1,4 +1,4 @@
-// GSD-2 + Durable subagent run status, tracking names, and result artifact store.
+// gsd-pi + Durable subagent run status, tracking names, and result artifact store.
 
 import * as fs from "node:fs";
 import * as path from "node:path";

package/src/resources/extensions/subagent/tests/launch.test.ts

@@ -1,4 +1,4 @@
-// GSD-2 + Subagent launch module regression tests.
+// gsd-pi + Subagent launch module regression tests.
 
 import assert from "node:assert/strict";
 import { mkdtempSync, rmSync } from "node:fs";

package/src/resources/extensions/subagent/tests/run-store.test.ts

@@ -1,4 +1,4 @@
-// GSD-2 + Subagent durable run-store tests.
+// gsd-pi + Subagent durable run-store tests.
 
 import assert from "node:assert/strict";
 import { mkdtempSync, rmSync } from "node:fs";

package/src/resources/extensions/visual-brief/artifact-policy.ts

@@ -1,4 +1,4 @@
-// GSD-2 + Visual Brief artifact policy
+// gsd-pi + Visual Brief artifact policy
 
 import { join } from "node:path";
 import { getAgentDir } from "@gsd/pi-coding-agent";

package/src/resources/extensions/visual-brief/index.ts

@@ -1,4 +1,4 @@
-// GSD-2 + Visual Brief bundled extension marker
+// gsd-pi + Visual Brief bundled extension marker
 
 import type { ExtensionAPI } from "@gsd/pi-coding-agent";
 

package/src/resources/extensions/visual-brief/page-contract.ts

@@ -1,4 +1,4 @@
-// GSD-2 + Visual Brief page contract
+// gsd-pi + Visual Brief page contract
 
 import type { VisualBriefMode } from "./prompts.js";
 

package/src/resources/extensions/visual-brief/prompts.ts

@@ -1,4 +1,4 @@
-// GSD-2 + Visual Brief prompt construction
+// gsd-pi + Visual Brief prompt construction
 
 import {
 	createVisualBriefArtifactPolicy,

package/src/resources/extensions/visual-brief/tests/visual-brief.test.ts

@@ -1,4 +1,4 @@
-// GSD-2 + Visual Brief command tests
+// gsd-pi + Visual Brief command tests
 
 import test from "node:test";
 import assert from "node:assert/strict";

package/src/resources/skills/api-design/SKILL.md

@@ -8,7 +8,7 @@ Shape an HTTP or GraphQL API so callers get predictable, evolvable, and honest s
 </objective>
 
 <context>
-GSD-2 has `design-an-interface` for general module-interface design; this skill is the HTTP/GraphQL specialization. REST and GraphQL carry baggage — status codes, verbs, nullability, pagination — that a generic interface-design discussion glosses over.
+gsd-pi has `design-an-interface` for general module-interface design; this skill is the HTTP/GraphQL specialization. REST and GraphQL carry baggage — status codes, verbs, nullability, pagination — that a generic interface-design discussion glosses over.
 
 Invocation points:
 - Adding a new public API endpoint

package/src/resources/skills/create-gsd-extension/SKILL.md

@@ -67,7 +67,7 @@ All domain knowledge in `references/`:
 **Infrastructure:** model-provider-management.md, remote-execution-overrides.md, packaging-distribution.md, mode-behavior.md
 **Spec:** `docs/extension-sdk/manifest-spec.md` — manifest format, tiers, validation
 **Testing:** `docs/extension-sdk/testing.md` — mock patterns, test conventions
-**SDK:** `docs/extension-sdk/` — the authoritative GSD-2 extension guide
+**SDK:** `docs/extension-sdk/` — the authoritative gsd-pi extension guide
 **Gotchas:** key-rules-gotchas.md
 </reference_index>
 

package/src/resources/skills/create-gsd-extension/templates/templates.test.ts

@@ -1,4 +1,4 @@
-// GSD-2 — Extension template import path validation
+// gsd-pi — Extension template import path validation
 // Copyright (c) 2026 Jeremy McSpadden <jeremy@fluxlabs.net>
 
 import { describe, it } from "node:test";

package/src/resources/skills/create-mcp-server/SKILL.md

@@ -8,7 +8,7 @@ Produce a high-quality MCP server that an LLM can actually use — not one that
 </objective>
 
 <context>
-GSD-2 consumes MCP heavily — see `src/resources/extensions/mcp-client/`, `src/resources/extensions/gsd/mcp-project-config.ts`, `src/resources/extensions/gsd/workflow-mcp.ts`, and `/gsd mcp` commands. Users frequently want to extend GSD with project-specific MCP servers (internal APIs, data sources, domain tools). This skill fills the authoring gap between "MCP exists" and "I have a working server."
+gsd-pi consumes MCP heavily — see `src/resources/extensions/mcp-client/`, `src/resources/extensions/gsd/mcp-project-config.ts`, `src/resources/extensions/gsd/workflow-mcp.ts`, and `/gsd mcp` commands. Users frequently want to extend GSD with project-specific MCP servers (internal APIs, data sources, domain tools). This skill fills the authoring gap between "MCP exists" and "I have a working server."
 
 Invocation points:
 - User describes a service or API they want an LLM to reach

package/src/resources/skills/create-skill/SKILL.md

@@ -88,9 +88,12 @@ Then proceed directly to the workflow.
 ## Skill Structure Quick Reference
 
 **Skill directories:**
+- Bundled/read-only: `~/.gsd/agent/skills/{skill-name}/`
 - Global: `~/.agents/skills/{skill-name}/`
 - Project-local: `.agents/skills/{skill-name}/`
 
+User-authored skills should target the global or project-local directories; GSD manages the bundled directory during install/update.
+
 **Simple skill (single file):**
 ```yaml
 ---

package/src/resources/skills/create-skill/references/gsd-skill-ecosystem.md

@@ -3,7 +3,12 @@ GSD-specific skill ecosystem details: directory conventions, discovery mechanics
 </overview>
 
 <skill_directories>
-GSD supports two skill directories, checked in order:
+GSD supports these skill directories, checked in order:
+
+**Bundled (GSD-managed):** `~/.gsd/agent/skills/`
+- Synced by GSD during install/update
+- Highest priority when names collide
+- Not the target for user-authored skills
 
 **User-scope (global):** `~/.agents/skills/`
 - Available in every GSD session regardless of working directory
@@ -14,17 +19,21 @@ GSD supports two skill directories, checked in order:
 - Committable to version control so team members share the same skill set
 - Ideal for project-specific workflows, deploy scripts, or conventions
 
-Skills in both directories follow the same SKILL.md format and router pattern conventions.
+**Claude Code compatibility:** `~/.claude/skills/` and `.claude/skills/`
+- Read after the standard Agent Skills directories
+- Useful for compatibility with existing Claude Code skill installs
+
+Skills in all directories follow the same SKILL.md format and router pattern conventions.
 </skill_directories>
 
 <skill_discovery>
 GSD auto-discovers skills at session start and during auto-mode:
 
-**Session start:** All skills in both directories are enumerated and their names + descriptions are injected into the system prompt as `<available_skills>`.
+**Session start:** All discovered skills are enumerated and their names + descriptions are injected into the system prompt as `<available_skills>`.
 
 **Auto-mode discovery:** `skill-discovery.ts` takes a snapshot of the skills directory at auto-mode start. On each unit boundary it diffs against the snapshot. Any new skills found are injected via a `<newly_discovered_skills>` XML block so the LLM sees them without requiring `/reload`.
 
-**Manual reload:** Running `/reload` re-scans both directories and updates the available skills list mid-session.
+**Manual reload:** Running `/reload` re-scans the skill directories and updates the available skills list mid-session.
 </skill_discovery>
 
 <skill_validation>

package/src/resources/skills/create-skill/workflows/create-new-skill.md

@@ -18,6 +18,8 @@
 1. **Global** (`~/.agents/skills/`) — Available in all GSD sessions
 2. **Project-local** (`.agents/skills/`) — Available only in this project
 
+Do not create user-authored skills under `~/.gsd/agent/skills/`; GSD owns that bundled skills directory.
+
 ## Step 2: Adaptive Requirements Gathering
 
 **If user provided context** (e.g., "build a skill for X"):

package/src/resources/skills/dependency-upgrade/SKILL.md

@@ -8,7 +8,7 @@ Turn a pile of outdated packages into a series of small, verifiable upgrades wit
 </objective>
 
 <context>
-GSD-2 ships a `/gsd start dep-upgrade` workflow template (`src/resources/extensions/gsd/workflow-templates/dep-upgrade.md`) that structures the phases: assess → upgrade → fix breaks → verify. This skill is the execution-level detail inside the upgrade phase — how to batch, how to verify, how to recover from a breaking upgrade without losing the good ones.
+gsd-pi ships a `/gsd start dep-upgrade` workflow template (`src/resources/extensions/gsd/workflow-templates/dep-upgrade.md`) that structures the phases: assess → upgrade → fix breaks → verify. This skill is the execution-level detail inside the upgrade phase — how to batch, how to verify, how to recover from a breaking upgrade without losing the good ones.
 
 Invocation points:
 - `/gsd start dep-upgrade` workflow is running

package/src/resources/skills/observability/SKILL.md

@@ -8,7 +8,7 @@ Instrument code so that a cold-start agent can understand what happened by readi
 </objective>
 
 <context>
-GSD-2's `VISION.md` lists "agent-first observability" as a principle, and the system prompt calls it out: "A future version of you will land in this codebase with no memory… you add observability because you're the one who'll need it at 3am." GSD-2 already exemplifies this — `activity/*.jsonl`, `journal/*.jsonl`, `metrics.json`, `doctor-history.jsonl` — but new code doesn't get that treatment automatically.
+gsd-pi's `VISION.md` lists "agent-first observability" as a principle, and the system prompt calls it out: "A future version of you will land in this codebase with no memory… you add observability because you're the one who'll need it at 3am." gsd-pi already exemplifies this — `activity/*.jsonl`, `journal/*.jsonl`, `metrics.json`, `doctor-history.jsonl` — but new code doesn't get that treatment automatically.
 
 This skill is the thinking process for adding it. Not "add logs everywhere" — add the *right* signals at the *right* decision points.
 
@@ -62,7 +62,7 @@ log.info({
 });
 ```
 
-Use the project's existing logger if one exists. In gsd-2, follow the patterns in `src/resources/extensions/gsd/activity-log.ts` and `src/resources/extensions/gsd/journal.ts` — structured JSONL, one event per line, with `ts`, `event`, and domain-specific fields.
+Use the project's existing logger if one exists. In gsd-pi, follow the patterns in `src/resources/extensions/gsd/activity-log.ts` and `src/resources/extensions/gsd/journal.ts` — structured JSONL, one event per line, with `ts`, `event`, and domain-specific fields.
 
 Avoid:
 - `console.log("here")` — what does "here" mean in six months?
@@ -86,14 +86,14 @@ await writeAtomically(
 );
 ```
 
-A fresh agent reading `.gsd/runtime/` sees what happened last, what was retried, and where the process stopped. Pattern exists already in gsd-2 — reuse the `atomic-write.ts` helpers and the `.gsd/runtime/` and `.gsd/forensics/` directories.
+A fresh agent reading `.gsd/runtime/` sees what happened last, what was retried, and where the process stopped. Pattern exists already in gsd-pi — reuse the `atomic-write.ts` helpers and the `.gsd/runtime/` and `.gsd/forensics/` directories.
 
 ## Step 4: Health and status surfaces
 
 For long-running processes:
 
 - **Health endpoint** (HTTP server) or **status file** (CLI tool). Cheap to call, no side effects. Returns current state: `{status: "healthy" | "degraded" | "down", ...diagnostics}`.
-- **Digest view** — a small representation of recent work. In gsd-2, this is `STATE.md` and the health widget. In a server, it's `/internal/status` with last 10 request summaries.
+- **Digest view** — a small representation of recent work. In gsd-pi, this is `STATE.md` and the health widget. In a server, it's `/internal/status` with last 10 request summaries.
 - **Minimal metrics** — counters for the 3–5 things that matter (requests, errors, active jobs). Not everything — just what drives alerts.
 
 Don't build a metrics empire. Build exactly what you'd check at 3am.

package/src/resources/skills/security-review/SKILL.md

@@ -8,7 +8,7 @@ Produce a security review that names specific exploit paths through the actual c
 </objective>
 
 <context>
-GSD-2's general `review` skill covers security as one of several categories. This skill is the deeper pass — triggered deliberately when security is the primary concern. It complements v1's `/gsd-secure-phase` concept, adapted to the gsd-2 artifact model.
+gsd-pi's general `review` skill covers security as one of several categories. This skill is the deeper pass — triggered deliberately when security is the primary concern. It complements v1's `/gsd-secure-phase` concept, adapted to the gsd-pi artifact model.
 
 Invocation points:
 - Any change touching authentication, authorization, session handling

package/dist/web/standalone/.next/static/chunks/app/layout-82017658b76e2494.js

@@ -1 +0,0 @@
-(self.webpackChunk_N_E=self.webpackChunk_N_E||[]).push([[7177],{20424:(e,r,s)=>{Promise.resolve().then(s.t.bind(s,66872,23)),Promise.resolve().then(s.bind(s,21942)),Promise.resolve().then(s.bind(s,61549))},21942:(e,r,s)=>{"use strict";s.d(r,{ThemeProvider:()=>n});var o=s(95155);s(12115);var t=s(70460);function n({children:e,...r}){return(0,o.jsx)(t.N,{...r,children:e})}},61549:(e,r,s)=>{"use strict";s.d(r,{Toaster:()=>a});var o=s(95155),t=s(70460),n=s(66609);let a=({...e})=>{let{theme:r="system"}=(0,t.D)();return(0,o.jsx)(n.l$,{theme:r,className:"toaster group",style:{"--normal-bg":"var(--popover)","--normal-text":"var(--popover-foreground)","--normal-border":"var(--border)"},...e})}},66872:()=>{}},e=>{e.O(0,[7513,4986,8441,3794,7358],()=>e(e.s=20424)),_N_E=e.O()}]);

package/dist/web/standalone/.next/static/chunks/app/page-d660442e70026068.js

@@ -1 +0,0 @@
-(self.webpackChunk_N_E=self.webpackChunk_N_E||[]).push([[8974],{5214:(e,t,n)=>{"use strict";Object.defineProperty(t,"__esModule",{value:!0}),Object.defineProperty(t,"workAsyncStorage",{enumerable:!0,get:function(){return r.workAsyncStorageInstance}});let r=n(17828)},17828:(e,t,n)=>{"use strict";Object.defineProperty(t,"__esModule",{value:!0}),Object.defineProperty(t,"workAsyncStorageInstance",{enumerable:!0,get:function(){return r}});let r=(0,n(64054).createAsyncLocalStorage)()},21957:(e,t,n)=>{"use strict";function r({moduleIds:e}){return null}Object.defineProperty(t,"__esModule",{value:!0}),Object.defineProperty(t,"PreloadChunks",{enumerable:!0,get:function(){return r}}),n(95155),n(47650),n(5214),n(2451),n(53887)},37206:(e,t,n)=>{"use strict";n.d(t,{default:()=>u.a});var r=n(75707),u=n.n(r)},41112:(e,t,n)=>{"use strict";function r({reason:e,children:t}){return t}Object.defineProperty(t,"__esModule",{value:!0}),Object.defineProperty(t,"BailoutToCSR",{enumerable:!0,get:function(){return r}}),n(1980)},64054:(e,t)=>{"use strict";Object.defineProperty(t,"__esModule",{value:!0});var n={bindSnapshot:function(){return s},createAsyncLocalStorage:function(){return a},createSnapshot:function(){return i}};for(var r in n)Object.defineProperty(t,r,{enumerable:!0,get:n[r]});let u=Object.defineProperty(Error("Invariant: AsyncLocalStorage accessed in runtime where it is not available"),"__NEXT_ERROR_CODE",{value:"E504",enumerable:!1,configurable:!0});class l{disable(){throw u}getStore(){}run(){throw u}exit(){throw u}enterWith(){throw u}static bind(e){return e}}let o="u">typeof globalThis&&globalThis.AsyncLocalStorage;function a(){return o?new o:new l}function s(e){return o?o.bind(e):l.bind(e)}function i(){return o?o.snapshot():function(e,...t){return e(...t)}}},66919:(e,t,n)=>{"use strict";n.r(t),n.d(t,{default:()=>l});var r=n(95155);let u=(0,n(37206).default)(()=>Promise.all([n.e(1838),n.e(4986),n.e(9370),n.e(8359)]).then(n.bind(n,18359)).then(e=>e.GSDAppShell),{loadableGenerated:{webpack:()=>[18359]},ssr:!1,loading:()=>(0,r.jsx)("div",{className:"flex h-screen items-center justify-center bg-background text-sm text-muted-foreground",children:"Loading workspace…"})});function l(){return(0,r.jsx)(u,{})}},68635:(e,t,n)=>{"use strict";Object.defineProperty(t,"__esModule",{value:!0}),Object.defineProperty(t,"default",{enumerable:!0,get:function(){return s}});let r=n(95155),u=n(12115),l=n(41112);function o(e){return{default:e&&"default"in e?e.default:e}}n(21957);let a={loader:()=>Promise.resolve(o(()=>null)),loading:null,ssr:!0},s=function(e){let t={...a,...e},n=(0,u.lazy)(()=>t.loader().then(o)),s=t.loading;function i(e){let o=s?(0,r.jsx)(s,{isLoading:!0,pastDelay:!0,error:null}):null,a=!t.ssr||!!t.loading,i=a?u.Suspense:u.Fragment,c=t.ssr?(0,r.jsxs)(r.Fragment,{children:[null,(0,r.jsx)(n,{...e})]}):(0,r.jsx)(l.BailoutToCSR,{reason:"next/dynamic",children:(0,r.jsx)(n,{...e})});return(0,r.jsx)(i,{...a?{fallback:o}:{},children:c})}return i.displayName="LoadableComponent",i}},75707:(e,t,n)=>{"use strict";Object.defineProperty(t,"__esModule",{value:!0}),Object.defineProperty(t,"default",{enumerable:!0,get:function(){return u}});let r=n(73623)._(n(68635));function u(e,t){let n={};"function"==typeof e&&(n.loader=e);let u={...n,...t};return(0,r.default)({...u,modules:u.loadableGenerated?.modules})}("function"==typeof t.default||"object"==typeof t.default&&null!==t.default)&&void 0===t.default.__esModule&&(Object.defineProperty(t.default,"__esModule",{value:!0}),Object.assign(t.default,t),e.exports=t.default)},87378:(e,t,n)=>{Promise.resolve().then(n.bind(n,66919))}},e=>{e.O(0,[8441,3794,7358],()=>e(e.s=87378)),_N_E=e.O()}]);

package/dist/web/standalone/.next/static/chunks/next/dist/client/components/builtin/global-error-1065c9a10c4c2187.js

@@ -1 +0,0 @@
-(self.webpackChunk_N_E=self.webpackChunk_N_E||[]).push([[9337],{86588:(e,s,_)=>{Promise.resolve().then(_.t.bind(_,27123,23))}},e=>{e.O(0,[8441,3794,7358],()=>e(e.s=86588)),_N_E=e.O()}]);