feat(task): added read-summarize override for subagents

- Parsed `read-summarize` frontmatter into `readSummarize` field.
- Applied `read.summarize.enabled: false` override on isolated subagent settings.
- Disabled summarization for `explore` and `librarian` agents.

docs/task-agent-discovery.md

@@ -24,7 +24,7 @@ It covers runtime behavior as implemented today, including precedence, invalid-d
 Task agents normalize into `AgentDefinition` (`src/task/types.ts`):
 
 - `name`, `description`, `systemPrompt` (required for a valid loaded agent)
-- optional `tools`, `spawns`, `model`, `thinkingLevel`, `output`, `blocking`
+- optional `tools`, `spawns`, `model`, `thinkingLevel`, `output`, `blocking`, `autoloadSkills`, `readSummarize`
 - `source`: `"bundled" | "user" | "project"`
 - optional `filePath`
 
@@ -35,6 +35,7 @@ Parsing comes from frontmatter via `parseAgentFields()` (`src/discovery/helpers.
 - `spawns` accepts `*`, CSV, or array
 - backward-compat behavior: if `spawns` missing but `tools` includes `task`, `spawns` becomes `*`
 - `output` is passed through as opaque schema data
+- `read-summarize: false` (parsed as `readSummarize`) forces the subagent's `read` tool to return verbatim file content instead of structural summaries — `runSubprocess` applies it as a `read.summarize.enabled: false` override on the subagent's isolated settings (`src/task/executor.ts`). `explore` and `librarian` ship with it disabled. Defaults to enabled when the field is absent.
 
 ## Bundled agents
 

packages/coding-agent/src/discovery/helpers.ts

@@ -212,6 +212,7 @@ export interface ParsedAgentFields {
 	output?: unknown;
 	thinkingLevel?: ThinkingLevel;
 	autoloadSkills?: string[];
+	readSummarize?: boolean;
 	blocking?: boolean;
 }
 
@@ -265,10 +266,11 @@ export function parseAgentFields(frontmatter: Record<string, unknown>): ParsedAg
 	const thinkingLevel = parseThinkingLevel(rawThinkingLevel);
 	const model = parseModelList(frontmatter.model);
 	const blocking = parseBoolean(frontmatter.blocking);
+	const readSummarize = parseBoolean(frontmatter.readSummarize);
 	const autoloadSkills = parseArrayOrCSV(frontmatter.autoloadSkills)
 		?.map(s => s.trim())
 		.filter(Boolean);
-	return { name, description, tools, spawns, model, output, thinkingLevel, blocking, autoloadSkills };
+	return { name, description, tools, spawns, model, output, thinkingLevel, blocking, autoloadSkills, readSummarize };
 }
 
 async function globIf(

packages/coding-agent/src/prompts/agents/explore.md

@@ -4,6 +4,7 @@ description: Fast read-only codebase scout returning compressed context for hand
 tools: read, search, find, web_search
 model: pi/smol
 thinking-level: med
+read-summarize: false
 output:
   properties:
     summary:

packages/coding-agent/src/prompts/agents/librarian.md

@@ -4,6 +4,7 @@ description: Researches external libraries and APIs by reading source code. Retu
 tools: read, search, find, bash, lsp, web_search, ast_grep
 model: pi/smol
 thinking-level: minimal
+read-summarize: false
 output:
   properties:
     answer:

packages/coding-agent/src/task/executor.ts

@@ -531,7 +531,10 @@ function createMCPProxyTools(mcpManager: MCPManager): CustomTool[] {
 	});
 }
 
-function createSubagentSettings(baseSettings: Settings): Settings {
+function createSubagentSettings(
+	baseSettings: Settings,
+	overrides?: Partial<Record<SettingPath, unknown>>,
+): Settings {
 	const snapshot: Partial<Record<SettingPath, unknown>> = {};
 	for (const key of Object.keys(SETTINGS_SCHEMA) as SettingPath[]) {
 		snapshot[key] = baseSettings.get(key);
@@ -545,6 +548,7 @@ function createSubagentSettings(baseSettings: Settings): Settings {
 		// the parent task approval is the authorization boundary. Use yolo mode
 		// to preserve unattended subagent execution. User `tools.approval` policies still apply.
 		"tools.approvalMode": "yolo",
+		...overrides,
 	});
 }
 
@@ -619,7 +623,10 @@ export async function runSubprocess(options: ExecutorOptions): Promise<SingleRes
 	}
 
 	const settings = options.settings ?? Settings.isolated();
-	const subagentSettings = createSubagentSettings(settings);
+	const subagentSettings = createSubagentSettings(
+		settings,
+		agent.readSummarize === false ? { "read.summarize.enabled": false } : undefined,
+	);
 	const maxRecursionDepth = settings.get("task.maxRecursionDepth") ?? 2;
 	const maxRuntimeMs = Math.max(0, Math.trunc(Number(settings.get("task.maxRuntimeMs") ?? 0) || 0));
 	const parentDepth = options.taskDepth ?? 0;

packages/coding-agent/src/task/types.ts

@@ -174,6 +174,8 @@ export interface AgentDefinition {
 	output?: unknown;
 	blocking?: boolean;
 	autoloadSkills?: string[];
+	/** When `false`, the agent's `read` tool returns verbatim file content instead of structural summaries. */
+	readSummarize?: boolean;
 	source: AgentSource;
 	filePath?: string;
 }

packages/coding-agent/test/discovery/agent-fields.test.ts

@@ -109,4 +109,27 @@ describe("parseAgentFields", () => {
 		expect(fields).toBeDefined();
 		expect(fields?.autoloadSkills).toBeUndefined();
 	});
+
+	test("parses readSummarize from boolean frontmatter", () => {
+		expect(parseAgentFields({ name: "explore", description: "desc", readSummarize: false })?.readSummarize).toBe(
+			false,
+		);
+		expect(parseAgentFields({ name: "explore", description: "desc", readSummarize: true })?.readSummarize).toBe(
+			true,
+		);
+	});
+
+	test("parses readSummarize from string frontmatter", () => {
+		expect(parseAgentFields({ name: "explore", description: "desc", readSummarize: "false" })?.readSummarize).toBe(
+			false,
+		);
+	});
+
+	test("ignores invalid readSummarize values", () => {
+		expect(parseAgentFields({ name: "explore", description: "desc", readSummarize: "nope" })?.readSummarize).toBeUndefined();
+	});
+
+	test("returns undefined readSummarize when field absent", () => {
+		expect(parseAgentFields({ name: "explore", description: "desc" })?.readSummarize).toBeUndefined();
+	});
 });