auto-mode 3.1.4 → 3.1.5

package/dist/cli.js

@@ -1,5 +1,5 @@
 #!/usr/bin/env node
-import { C as pickJudgeLane, E as updateJSONFile, S as CLI_LIMITS, T as resolveXDGDir, _ as readCheckout, a as resolveSessionScopePath, b as loadConfig, c as splitGitCommand, d as loadCheckoutFacts, f as writeActionDiagnostic, g as isUnresolvedWord, h as resolveShellPath, i as classifyAction, l as loadPolicy, m as collectShellCommands, n as readRuntime, o as loadSessionScope, p as toRepositorySlug, s as EMPTY_SESSION_SCOPE, t as parseActionRequest, u as resolveShippedPolicyDir, v as readGitEntry, w as M, x as runCapped, y as DEFAULT_SCOPE_SOURCES } from "./parse-action-request-DlsgKqiF.js";
+import { C as pickJudgeLane, E as updateJSONFile, S as CLI_LIMITS, T as resolveXDGDir, _ as readCheckout, a as resolveSessionScopePath, b as loadConfig, c as splitGitCommand, d as loadCheckoutFacts, f as writeActionDiagnostic, g as isUnresolvedWord, h as resolveShellPath, i as classifyAction, l as loadPolicy, m as collectShellCommands, n as readRuntime, o as loadSessionScope, p as toRepositorySlug, s as EMPTY_SESSION_SCOPE, t as parseActionRequest, u as resolveShippedPolicyDir, v as readGitEntry, w as M, x as runCapped, y as DEFAULT_SCOPE_SOURCES } from "./parse-action-request-CHuF6LYV.js";
 import { a as literal, d as string, i as discriminatedUnion, l as record, p as unknown, r as boolean, s as number, u as strictObject } from "./schemas-B80Yxv6m.js";
 import { createHash, randomUUID } from "node:crypto";
 import { chmod, mkdir, open, readFile, realpath, stat } from "node:fs/promises";

package/dist/index.d.ts

@@ -147,10 +147,10 @@ type MessageOrigin = (typeof MESSAGE_ORIGINS)[number];
 interface UserTask {
   readonly text: string;
   readonly origin: MessageOrigin;
+  readonly position?: 'opening' | undefined;
 }
 interface DirectUserMessage extends UserTask {
   readonly freshness?: 'stale' | undefined;
-  readonly position?: 'opening' | undefined;
 }
 interface DelegatedTask {
   readonly text: string;

package/dist/index.js

@@ -1,4 +1,4 @@
-import { b as loadConfig$1, i as classifyAction$1, n as readRuntime, r as readHostEnvironment, t as parseActionRequest } from "./parse-action-request-DlsgKqiF.js";
+import { b as loadConfig$1, i as classifyAction$1, n as readRuntime, r as readHostEnvironment, t as parseActionRequest } from "./parse-action-request-CHuF6LYV.js";
 //#region src/index.ts
 function classifyAction(request, config, options = {}) {
 	return classifyAction$1(request, config, {

package/dist/{parse-action-request-DlsgKqiF.js → parse-action-request-CHuF6LYV.js}

@@ -251,6 +251,25 @@ async function tryUpdateDenialBudget(stderr, path, plan) {
 	}
 }
 //#endregion
+//#region src/scope/find-reply-channel.ts
+function findReplyChannel(tool, input, channels) {
+	return channels.find((channel) => channel.tool === tool && input[channel.field] === channel.recipient) ?? null;
+}
+//#endregion
+//#region src/bypass/classify-reply-call.ts
+async function classifyReplyCall(call, scope, hasConfiguredDenies) {
+	const channel = scope === null ? null : findReplyChannel(call.toolName, call.toolInput, scope.replyChannels);
+	if (channel === null) return { kind: "none" };
+	if ((await import("./classify-reply-message-CdHLlNXA.js")).classifyReplyMessage(call.toolInput).kind !== "bypass") return { kind: "none" };
+	return hasConfiguredDenies ? {
+		kind: "exempt",
+		rules: ["Outbound Communication"]
+	} : {
+		kind: "bypass",
+		channel
+	};
+}
+//#endregion
 //#region src/state/resolve-xdg-dir.ts
 const BASE_DIRS = {
 	config: {
@@ -1098,11 +1117,6 @@ const SCRATCH_PATHS = [
 	"/dev/stderr"
 ];
 //#endregion
-//#region src/scope/find-reply-channel.ts
-function findReplyChannel(tool, input, channels) {
-	return channels.find((channel) => channel.tool === tool && input[channel.field] === channel.recipient) ?? null;
-}
-//#endregion
 //#region src/shell/is-unresolved-word.ts
 function isUnresolvedWord(word) {
 	return /[$`]/u.test(word);
@@ -3091,7 +3105,7 @@ async function writeActionDiagnostic(payload, entry, host, warnings) {
 	const agentKey = agentID === null ? payload.sessionID : `${payload.sessionID}\0${agentID}`;
 	const isReviewed = entry.judge !== void 0 || entry.grant !== void 0;
 	const record = {
-		schemaVersion: 8,
+		schemaVersion: 9,
 		time: (/* @__PURE__ */ new Date()).toISOString(),
 		invocationID: entry.invocationID,
 		sessionHash: toHash(payload.sessionID),
@@ -3105,7 +3119,8 @@ async function writeActionDiagnostic(payload, entry, host, warnings) {
 		diagnostics: entry.diagnostics ?? null,
 		judge: entry.judge === void 0 ? null : buildLoggedReview(entry.judge),
 		grant: entry.grant === void 0 ? null : buildLoggedReview(entry.grant),
-		action: isReviewed ? buildLoggedAction(payload) : null
+		action: isReviewed ? buildLoggedAction(payload) : null,
+		taskEvidence: buildTaskEvidence(entry.taskContext ?? payload.decisionContext)
 	};
 	try {
 		await mkdir(dirname(path), {
@@ -3159,6 +3174,15 @@ function hasSecret(value) {
 	if (typeof value === "object" && value !== null) return Object.values(value).some((item) => hasSecret(item));
 	return false;
 }
+function buildTaskEvidence(context) {
+	if (context === void 0) return null;
+	const message = context.lastDirectUserMessage;
+	return {
+		userMessage: message === null ? null : message.freshness ?? "current",
+		originalUserTask: context.originalUserTask?.origin ?? null,
+		delegatedTask: context.delegatedTask !== null
+	};
+}
 //#endregion
 //#region src/diagnostics/write-judge-usage.ts
 async function writeJudgeUsage(usage, host, warnings) {
@@ -3946,6 +3970,31 @@ async function classifyWithJudge(payload, evidence, judge, options) {
 	});
 }
 //#endregion
+//#region src/model/build-budget-omitted-context.ts
+function buildBudgetOmittedContext(context, field) {
+	return {
+		...context,
+		[field]: null,
+		omittedTaskContext: [...context.omittedTaskContext.filter((item) => item.field !== field), {
+			field,
+			reason: "budget"
+		}]
+	};
+}
+//#endregion
+//#region src/model/build-task-context.ts
+function buildTaskContext(context) {
+	let result = {
+		...context,
+		omittedTaskContext: [...context.omittedTaskContext]
+	};
+	for (const field of ["originalUserTask", "delegatedTask"]) {
+		const task = result[field];
+		if (task !== null && Buffer.byteLength(task.text) > 4096) result = buildBudgetOmittedContext(result, field);
+	}
+	return result;
+}
+//#endregion
 //#region src/model/build-task-scope-summary.ts
 function buildTaskScopeSummary(scope) {
 	return {
@@ -4230,31 +4279,6 @@ function resolveHooksPath(path, worktree, home) {
 	return path.startsWith("~/") ? join(home, path.slice(2)) : resolve(worktree, path);
 }
 //#endregion
-//#region src/model/build-budget-omitted-context.ts
-function buildBudgetOmittedContext(context, field) {
-	return {
-		...context,
-		[field]: null,
-		omittedTaskContext: [...context.omittedTaskContext.filter((item) => item.field !== field), {
-			field,
-			reason: "budget"
-		}]
-	};
-}
-//#endregion
-//#region src/model/build-task-context.ts
-function buildTaskContext(context) {
-	let result = {
-		...context,
-		omittedTaskContext: [...context.omittedTaskContext]
-	};
-	for (const field of ["originalUserTask", "delegatedTask"]) {
-		const task = result[field];
-		if (task !== null && Buffer.byteLength(task.text) > 4096) result = buildBudgetOmittedContext(result, field);
-	}
-	return result;
-}
-//#endregion
 //#region src/git/parse-config-parameters.ts
 function parseConfigParameters(text) {
 	const parameters = [];
@@ -4798,6 +4822,7 @@ function buildDecisionRequest(payload, policy, options) {
 	const rulesSource = options.rulesSource ?? "replacement";
 	const repositoryContext = options.repositoryContext ?? null;
 	const mcpServers = options.mcpServers ?? [];
+	const exemptRules = options.exemptRules ?? [];
 	const rules = {};
 	let tier = null;
 	for (const section of policy.split(/(?=^## |^### )/mu)) if (section.startsWith("## HARD BLOCK rules")) tier = "hard";
@@ -4808,6 +4833,7 @@ function buildDecisionRequest(payload, policy, options) {
 		const end = newline === -1 ? section.length : newline;
 		const name = section.slice(4, end).trim();
 		if (name === "" || Object.values(rules).some((rule) => rule.name === name)) throw new Error("Policy has an empty or duplicate rule");
+		if (rulesSource === "shipped" && exemptRules.includes(name)) continue;
 		rules[`rule_${Object.keys(rules).length}`] = {
 			name,
 			tier,
@@ -5033,6 +5059,30 @@ function classifyFailure(attempt, timeoutMs) {
 	};
 }
 //#endregion
+//#region src/model/build-sized-decision-input.ts
+function buildSizedDecisionInput(model, request) {
+	return buildFittedInput(buildFittedInput(buildInput(model, request.state, request.questions), "delegatedTask"), "originalUserTask");
+}
+function buildInput(model, state, questions) {
+	const input = {
+		model,
+		state,
+		questions
+	};
+	return {
+		input,
+		body: JSON.stringify(input)
+	};
+}
+function buildFittedInput(sized, field) {
+	const context = sized.input.state.taskContext;
+	if (Buffer.byteLength(sized.body) <= 1e5 || context === void 0 || context[field] === null) return sized;
+	return buildInput(sized.input.model, {
+		...sized.input.state,
+		taskContext: buildBudgetOmittedContext(context, field)
+	}, sized.input.questions);
+}
+//#endregion
 //#region src/model/collect-config-file-paths.ts
 function collectConfigFilePaths(command) {
 	const values = splitShellCommand(command).segments.flatMap((segment) => splitShellTokens(segment.text).tokens.flatMap((token) => toWordValues(token))).flatMap((word) => {
@@ -5069,18 +5119,7 @@ function buildDecisionResponseSchema(choices) {
 //#endregion
 //#region src/model/send-decision.ts
 async function sendDecision(provider, apiKey, request, options) {
-	const input = {
-		model: provider.model,
-		state: { ...request.state },
-		questions: request.questions
-	};
-	let body = JSON.stringify(input);
-	for (const field of ["delegatedTask", "originalUserTask"]) {
-		const context = input.state.taskContext;
-		if (Buffer.byteLength(body) <= 1e5 || context === void 0 || context[field] === null) continue;
-		input.state.taskContext = buildBudgetOmittedContext(context, field);
-		body = JSON.stringify(input);
-	}
+	const body = buildSizedDecisionInput(provider.model, request).body;
 	const requestBytes = Buffer.byteLength(body);
 	if (requestBytes > 1e5) throw new DecisionRequestError("request-too-large", requestBytes, "Decision input exceeds 100000 bytes; refusing to truncate the action or user message");
 	let response;
@@ -5165,7 +5204,8 @@ async function classifyWithJev(payload, config, options) {
 			repositoryContext,
 			mcpServers,
 			localHookConfig,
-			hookFacts
+			hookFacts,
+			exemptRules: options.exemptRules
 		});
 		const signal = createDeadlineSignal(config.provider.timeoutMs, options);
 		if (signal === null) throw new DOMException("Evaluation deadline expired", "AbortError");
@@ -5175,6 +5215,7 @@ async function classifyWithJev(payload, config, options) {
 			fetch: runtime.fetch
 		});
 		stage = "response";
+		options.onTaskContext?.(buildSizedDecisionInput(config.provider.model, request).input.state.taskContext);
 		return buildJevOutcome({
 			kind: "reply",
 			request,
@@ -5981,31 +6022,65 @@ async function probeLauncherSpawner(location, readTextFile) {
 //#region src/scope/resolve-opening-prompt.ts
 async function resolveOpeningPrompt(request, runtime) {
 	const context = request.decisionContext;
-	const message = context?.lastDirectUserMessage;
-	if (context === void 0 || message?.position !== "opening") return request;
-	const { position: _, ...unmarked } = message;
+	const task = context?.originalUserTask ?? null;
+	const message = context?.lastDirectUserMessage ?? null;
+	const opening = task?.position === "opening" ? task : message;
+	if (context === void 0 || opening?.position !== "opening") return request;
 	const isSpawned = await probeLauncherSpawner(runtime.host.env["AUTO_MODE_SCOPE_FILE"], runtime.readTextFile);
 	return {
 		...request,
-		decisionContext: {
+		decisionContext: isSpawned ? buildSpawnedContext(context, opening.text) : {
 			...context,
-			lastDirectUserMessage: isSpawned ? null : unmarked
+			originalUserTask: task === null ? null : toUnmarked(task),
+			lastDirectUserMessage: message === null ? null : toUnmarked(message)
 		}
 	};
 }
+function buildSpawnedContext(context, openingText) {
+	const isTaskOpening = context.originalUserTask?.position === "opening";
+	const message = context.lastDirectUserMessage;
+	const isOmitted = context.omittedTaskContext.some((omission) => omission.field === "originalUserTask");
+	return {
+		...context,
+		originalUserTask: isTaskOpening ? null : context.originalUserTask,
+		delegatedTask: context.agentID === null ? context.delegatedTask ?? {
+			text: openingText,
+			origin: "agent.spawn"
+		} : context.delegatedTask,
+		lastDirectUserMessage: message?.position === "opening" ? null : message,
+		omittedTaskContext: isTaskOpening && !isOmitted ? [...context.omittedTaskContext, {
+			field: "originalUserTask",
+			reason: "unavailable"
+		}] : context.omittedTaskContext
+	};
+}
+function toUnmarked(value) {
+	const { position: _, ...unmarked } = value;
+	return unmarked;
+}
 //#endregion
 //#region src/classify-action.ts
 async function classifyAction(request, config, options) {
-	const stderr = options.stderr;
-	const host = options.runtime.host;
 	const invocationID = randomUUID();
-	if (options.actionLog === true) await writeActionDiagnostic(request, {
-		invocationID,
-		status: "started"
-	}, host, stderr);
 	const resolved = await resolveOpeningPrompt(request, options.runtime);
-	const outcome = options.budget === true ? await classifyWithBudget(resolved, config, options) : await classifyStages(resolved, config, options);
-	if (options.actionLog === true) await writeActionDiagnostic(request, {
+	const offered = resolved.decisionContext === void 0 ? void 0 : buildTaskContext(resolved.decisionContext);
+	if (options.actionLog === true) await writeActionDiagnostic(resolved, {
+		invocationID,
+		status: "started",
+		taskContext: offered
+	}, options.runtime.host, options.stderr);
+	const received = { context: offered };
+	const onTaskContext = (context) => {
+		received.context = context;
+	};
+	const outcome = options.budget === true ? await classifyWithBudget(resolved, config, {
+		...options,
+		onTaskContext
+	}) : await classifyStages(resolved, config, {
+		...options,
+		onTaskContext
+	});
+	if (options.actionLog === true) await writeActionDiagnostic(resolved, {
 		invocationID,
 		status: outcome.status,
 		verdict: outcome.verdict?.kind ?? "defer",
@@ -6014,8 +6089,9 @@ async function classifyAction(request, config, options) {
 		...outcome.escalation === void 0 ? {} : { escalation: outcome.escalation },
 		...outcome.diagnostics === void 0 ? {} : { diagnostics: outcome.diagnostics },
 		...outcome.judge === void 0 ? {} : { judge: outcome.judge },
-		...outcome.grant === void 0 ? {} : { grant: outcome.grant }
-	}, host, stderr);
+		...outcome.grant === void 0 ? {} : { grant: outcome.grant },
+		taskContext: received.context
+	}, options.runtime.host, options.stderr);
 	return outcome;
 }
 async function classifyWithBudget(request, config, options) {
@@ -6052,7 +6128,8 @@ async function classifyStages(request, config, options) {
 		local: { kind: "escalate" },
 		checkout: null,
 		scope: null,
-		grant: null
+		grant: null,
+		exemptRules: []
 	};
 	const reviewed = await classifyWithGrant(ctx, await runStages(ctx));
 	if (reviewed.verdict?.kind !== "deny") return reviewed;
@@ -6236,9 +6313,12 @@ async function classifyEditBypass(ctx) {
 	};
 }
 async function classifyReplyChannel(ctx) {
-	const channel = ctx.hasConfiguredDenies || ctx.scope === null ? null : findReplyChannel(ctx.request.toolName, ctx.request.toolInput, ctx.scope.replyChannels);
-	if (channel === null) return { kind: "pass" };
-	if ((await import("./classify-reply-message-CdHLlNXA.js")).classifyReplyMessage(ctx.request.toolInput).kind !== "bypass") return { kind: "pass" };
+	const reply = await classifyReplyCall(ctx.request, ctx.scope, ctx.hasConfiguredDenies);
+	if (reply.kind !== "bypass") return {
+		kind: "pass",
+		facts: { exemptRules: reply.kind === "exempt" ? reply.rules : [] }
+	};
+	const channel = reply.channel;
 	return {
 		kind: "decided",
 		outcome: {
@@ -6267,7 +6347,8 @@ async function classifyJev(ctx) {
 		taskScope: ctx.scope === null ? void 0 : buildTaskScopeSummary(ctx.scope),
 		claudeSettings: ctx.claudeSettings,
 		checkout: ctx.checkout ?? void 0,
-		readCheckout: ctx.readCheckout
+		readCheckout: ctx.readCheckout,
+		exemptRules: ctx.exemptRules
 	});
 	return {
 		...outcome,
@@ -6359,12 +6440,12 @@ const userTaskSchema = strictObject({
 		"composer",
 		"bridge",
 		"sdk"
-	])
+	]),
+	position: literal("opening").optional()
 });
 const directUserMessageSchema = strictObject({
 	...userTaskSchema.shape,
-	freshness: literal("stale").optional(),
-	position: literal("opening").optional()
+	freshness: literal("stale").optional()
 });
 const delegatedTaskSchema = strictObject({
 	text: string(),

package/mods/auto-mode/contract/types.ts

@@ -5,11 +5,11 @@ export type MessageOrigin = (typeof MESSAGE_ORIGINS)[number];
 export interface UserTask {
   readonly text: string;
   readonly origin: MessageOrigin;
+  readonly position?: 'opening' | undefined;
 }
 
 export interface DirectUserMessage extends UserTask {
   readonly freshness?: 'stale' | undefined;
-  readonly position?: 'opening' | undefined;
 }
 
 interface DelegatedTask {

package/mods/auto-mode/hooks/build-prompt-context.ts

@@ -36,13 +36,16 @@ export function buildPromptContext(
     };
   }
 
-  const message: UserTask = { text: input.text, origin };
   const isOpeningPrompt = previous?.awaitsOpeningPrompt ?? true;
 
+  const message: UserTask = isOpeningPrompt
+    ? { text: input.text, origin, position: 'opening' }
+    : { text: input.text, origin };
+
   return {
     originalUserTask:
       previous?.originalUserTask ?? (previous?.canCaptureOriginal === true ? message : null),
-    lastDirectUserMessage: isOpeningPrompt ? { ...message, position: 'opening' } : message,
+    lastDirectUserMessage: message,
     canCaptureOriginal: previous?.canCaptureOriginal ?? false,
     awaitsOpeningPrompt: false,
   };

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "auto-mode",
-  "version": "3.1.4",
+  "version": "3.1.5",
   "description": "A permission classifier library and Claude Code mod, so agents that are not Claude can run unattended",
   "keywords": [
     "agent",
@@ -52,6 +52,7 @@
     "eval:question-shape": "bun --no-env-file --conditions=auto-mode-eval evals/runners/run-question-shape-evaluation.ts",
     "eval:relay-consent": "bun --no-env-file --conditions=auto-mode-eval evals/runners/run-relay-consent-evaluation.ts",
     "eval:second-judge": "bun --no-env-file --conditions=auto-mode-eval evals/runners/run-second-judge-evaluation.ts",
+    "eval:spawned-task": "bun --no-env-file --conditions=auto-mode-eval evals/runners/run-spawned-task-evaluation.ts",
     "eval:stale-consent": "bun --no-env-file --conditions=auto-mode-eval evals/runners/run-stale-consent-evaluation.ts",
     "format": "oxfmt . && format-codemod --quiet .",
     "format:check": "oxfmt --check . && format-codemod --check --quiet .",