Merge hooks and custom-tools into unified extensions system (#454)

Breaking changes:
- Settings: 'hooks' and 'customTools' arrays replaced with 'extensions'
- CLI: '--hook' and '--tool' flags replaced with '--extension' / '-e'
- API: HookMessage renamed to CustomMessage, role 'hookMessage' to 'custom'
- API: FileSlashCommand renamed to PromptTemplate
- API: discoverSlashCommands() renamed to discoverPromptTemplates()
- Directories: commands/ renamed to prompts/ for prompt templates

Migration:
- Session version bumped to 3 (auto-migrates v2 sessions)
- Old 'hookMessage' role entries converted to 'custom'

Structural changes:
- src/core/hooks/ and src/core/custom-tools/ merged into src/core/extensions/
- src/core/slash-commands.ts renamed to src/core/prompt-templates.ts
- examples/hooks/ and examples/custom-tools/ merged into examples/extensions/
- docs/hooks.md and docs/custom-tools.md merged into docs/extensions.md

New test coverage:
- test/extensions-runner.test.ts (10 tests)
- test/extensions-discovery.test.ts (26 tests)
- test/prompt-templates.test.ts

packages/coding-agent/examples/extensions/custom-compaction.ts

@@ -0,0 +1,114 @@
+/**
+ * Custom Compaction Extension
+ *
+ * Replaces the default compaction behavior with a full summary of the entire context.
+ * Instead of keeping the last 20k tokens of conversation turns, this extension:
+ * 1. Summarizes ALL messages (messagesToSummarize + turnPrefixMessages)
+ * 2. Discards all old turns completely, keeping only the summary
+ *
+ * This example also demonstrates using a different model (Gemini Flash) for summarization,
+ * which can be cheaper/faster than the main conversation model.
+ *
+ * Usage:
+ *   pi --extension examples/extensions/custom-compaction.ts
+ */
+
+import { complete, getModel } from "@mariozechner/pi-ai";
+import type { ExtensionAPI } from "@mariozechner/pi-coding-agent";
+import { convertToLlm, serializeConversation } from "@mariozechner/pi-coding-agent";
+
+export default function (pi: ExtensionAPI) {
+	pi.on("session_before_compact", async (event, ctx) => {
+		ctx.ui.notify("Custom compaction extension triggered", "info");
+
+		const { preparation, branchEntries: _, signal } = event;
+		const { messagesToSummarize, turnPrefixMessages, tokensBefore, firstKeptEntryId, previousSummary } = preparation;
+
+		// Use Gemini Flash for summarization (cheaper/faster than most conversation models)
+		const model = getModel("google", "gemini-2.5-flash");
+		if (!model) {
+			ctx.ui.notify(`Could not find Gemini Flash model, using default compaction`, "warning");
+			return;
+		}
+
+		// Resolve API key for the summarization model
+		const apiKey = await ctx.modelRegistry.getApiKey(model);
+		if (!apiKey) {
+			ctx.ui.notify(`No API key for ${model.provider}, using default compaction`, "warning");
+			return;
+		}
+
+		// Combine all messages for full summary
+		const allMessages = [...messagesToSummarize, ...turnPrefixMessages];
+
+		ctx.ui.notify(
+			`Custom compaction: summarizing ${allMessages.length} messages (${tokensBefore.toLocaleString()} tokens) with ${model.id}...`,
+			"info",
+		);
+
+		// Convert messages to readable text format
+		const conversationText = serializeConversation(convertToLlm(allMessages));
+
+		// Include previous summary context if available
+		const previousContext = previousSummary ? `\n\nPrevious session summary for context:\n${previousSummary}` : "";
+
+		// Build messages that ask for a comprehensive summary
+		const summaryMessages = [
+			{
+				role: "user" as const,
+				content: [
+					{
+						type: "text" as const,
+						text: `You are a conversation summarizer. Create a comprehensive summary of this conversation that captures:${previousContext}
+
+1. The main goals and objectives discussed
+2. Key decisions made and their rationale
+3. Important code changes, file modifications, or technical details
+4. Current state of any ongoing work
+5. Any blockers, issues, or open questions
+6. Next steps that were planned or suggested
+
+Be thorough but concise. The summary will replace the ENTIRE conversation history, so include all information needed to continue the work effectively.
+
+Format the summary as structured markdown with clear sections.
+
+<conversation>
+${conversationText}
+</conversation>`,
+					},
+				],
+				timestamp: Date.now(),
+			},
+		];
+
+		try {
+			// Pass signal to honor abort requests (e.g., user cancels compaction)
+			const response = await complete(model, { messages: summaryMessages }, { apiKey, maxTokens: 8192, signal });
+
+			const summary = response.content
+				.filter((c): c is { type: "text"; text: string } => c.type === "text")
+				.map((c) => c.text)
+				.join("\n");
+
+			if (!summary.trim()) {
+				if (!signal.aborted) ctx.ui.notify("Compaction summary was empty, using default compaction", "warning");
+				return;
+			}
+
+			// Return compaction content - SessionManager adds id/parentId
+			// Use firstKeptEntryId from preparation to keep recent messages
+			return {
+				compaction: {
+					summary,
+					firstKeptEntryId,
+					tokensBefore,
+				},
+			};
+		} catch (error) {
+			const message = error instanceof Error ? error.message : String(error);
+			ctx.ui.notify(`Compaction failed: ${message}`, "error");
+			// Fall back to default compaction on error
+			return;
+		}
+	});
+}