@gitroll/tacitt-cli 0.6.10 → 0.6.12

package/skills/tacitt/scripts/privacy-validator.mjs

@@ -5484,12 +5484,33 @@ object({
 });
 //#endregion
 //#region packages/processing/src/errors.ts
+const sitePattern = /^[a-z][a-z0-9]*(?:-[a-z0-9]+){0,7}$/;
+const harnessPattern = /^[a-z]{1,16}$/;
+const versionPattern = /^\d{1,6}(?:\.\d{1,6}){1,3}(?:-[0-9A-Za-z.]{1,24})?$/;
+/** The validated site fields of `value`; anything that does not match its pattern is dropped. */
+function siteFields(value) {
+	if (typeof value !== "object" || value === null) return void 0;
+	const site = Reflect.get(value, "site");
+	if (typeof site !== "string" || !sitePattern.test(site)) return void 0;
+	const harness = Reflect.get(value, "harness");
+	const harnessVersion = Reflect.get(value, "harnessVersion");
+	return {
+		site,
+		...typeof harness === "string" && harnessPattern.test(harness) ? { harness } : {},
+		...typeof harnessVersion === "string" && versionPattern.test(harnessVersion) ? { harnessVersion } : {}
+	};
+}
 var ProcessingError = class extends Error {
 	code;
+	site;
+	harness;
+	harnessVersion;
 	constructor(code, options) {
 		super(code, options);
 		this.name = "ProcessingError";
 		this.code = code;
+		const named = siteFields(options);
+		if (named) Object.assign(this, named);
 	}
 };
 function fail(code, options) {
@@ -5552,6 +5573,10 @@ const credentialNames = Object.freeze([
 	"auth_token",
 	"session_token",
 	"authorization",
+	"authorization_code",
+	"oauth_code",
+	"密碼",
+	"密码",
 	"proxy_authorization",
 	"x_api_key",
 	"private_key",
@@ -5837,13 +5862,13 @@ function assignments(input, output) {
 	}
 	if (input.field.kind === "key") return;
 	const consumed = headers(input, output);
-	if (!/[:=]|--/u.test(input.text)) return;
+	if (!/[:=：]|--/u.test(input.text)) return;
 	if (!mayContainCredentialName(input.text)) return;
-	const expression = /(?:(?<![\p{L}\p{N}_/?&#-])(\\?["']?)([\p{L}\p{N}_-]{1,64})\1([ \t]*)[:=]([ \t]*)|(?<![\p{L}\p{N}_-])--([\p{L}\p{N}_-]{1,64})([ \t]+)(?![ \t-]))/giu;
+	const expression = /(?:(?<![\p{L}\p{N}_/?&#-])(\\?["']?)((?:oauth[ \t]+)?authorization[ \t]+code|oauth[ \t]+code|[\p{L}\p{N}_-]{1,64})\1([ \t]*)[:=：]([ \t]*)|(?<![\p{L}\p{N}_-])--([\p{L}\p{N}_-]{1,64})([ \t]+)(?![ \t-]))/giu;
 	let match;
 	let headerIndex = 0;
 	while ((match = expression.exec(input.text)) !== null) {
-		const label = normalizeCredentialName(match[2] ?? match[5] ?? "");
+		const label = normalizeCredentialName((match[2] ?? match[5] ?? "").replace(/[ \t]+/gu, "_"));
 		if (!credentialName(label)) continue;
 		while ((consumed[headerIndex]?.end ?? Infinity) <= match.index) headerIndex += 1;
 		const header = consumed[headerIndex];
@@ -7661,10 +7686,10 @@ function urlActions(locator, input, result) {
 			actions: [action]
 		});
 	}
-	const inspect = (range, key, query = false) => {
+	const inspect = (range, key, query = false, oauthResponse = false) => {
 		if (range.start >= range.end) return;
 		const raw = input.text.slice(range.start, range.end);
-		const credential = key !== void 0 && credentialName(key);
+		const credential = key !== void 0 && (credentialName(key) || oauthResponse && key === "code");
 		if (!withinLimit({
 			raw,
 			range
@@ -7711,6 +7736,8 @@ function urlActions(locator, input, result) {
 	const end = hash < 0 ? locator.raw.length : hash;
 	const question = locator.raw.indexOf("?", pathEnd);
 	if (question >= 0 && question < end) {
+		const keys = new URLSearchParams(locator.raw.slice(question + 1, end));
+		const oauthResponse = (locator.scheme === "http" || locator.scheme === "https") && keys.has("code") && (keys.has("state") || keys.has("session_state") || /(?:^|\/)(?:oauth2?|callback|oauth2?[_-]callback)(?:\/|$)/iu.test(locator.raw.slice(suffix, pathEnd)));
 		start = question + 1;
 		for (let index = start; index <= end; index++) {
 			if (index !== end && locator.raw[index] !== "&") continue;
@@ -7720,7 +7747,7 @@ function urlActions(locator, input, result) {
 				if (key !== null) inspect({
 					start: offset + equals + 1,
 					end: offset + index
-				}, key, true);
+				}, key, true, oauthResponse);
 			} else inspect({
 				start: offset + start,
 				end: offset + index
@@ -9431,7 +9458,7 @@ Object.freeze([
 ]);
 Object.freeze({
 	implementation: originalInspectionVersion,
-	detector: `static-catalog-11`,
+	detector: `static-catalog-12`,
 	parser: nativeParserVersion,
 	dictionary: entityDictionaryVersion
 });
@@ -9679,6 +9706,24 @@ const termEvidenceSchema = object({
 	rationale: termAssessmentSchema.shape.rationale,
 	stance: termAssessmentSchema.shape.stance
 }).strict().refine((evidence) => evidence.end > evidence.start);
+const namedBatch = object({
+	journeyId: id$1,
+	batchId: id$1,
+	batchDigest: digest$2
+}).strict();
+discriminatedUnion("kind", [namedBatch.extend({ kind: literal("inspect") }).strict(), namedBatch.extend({
+	kind: literal("apply"),
+	expectedPackId: id$1,
+	idempotencyKey: id$1
+}).strict()]);
+object({
+	status: _enum([
+		"eligible",
+		"reopened",
+		"replayed"
+	]),
+	packId: id$1
+}).strict();
 object({
 	journeyId: id$1,
 	idempotencyKey: id$1,
@@ -9688,6 +9733,10 @@ object({
 	units: array(object({
 		batchId: id$1,
 		batchDigest: digest$2,
+		coverage: object({
+			attemptId: id$1,
+			receipts: array(digest$2).min(1).max(8e4)
+		}).strict().optional(),
 		/** Legacy unit summary in stored packs; accepted and never read (ADR 0178). */
 		summary: string().trim().min(1).refine((value) => Buffer.byteLength(value) <= 1024).optional(),
 		exclude: string().trim().refine((value) => codePoints(value) >= 1 && codePoints(value) <= 500).nullable()
@@ -10107,6 +10156,13 @@ const journeyArtifactSchema = object({
 		archiveBytes: count$1
 	}).strict()
 }).strict();
+/** A closed reader attempt that left named units without a coverage receipt. */
+const reviewInterruptionSchema = object({
+	inputRevision: count$1.min(1),
+	packId: journeyIdSchema,
+	reason: _enum(["reader-refused", "retry-exhausted"]),
+	batchIds: array(journeyIdSchema).min(1).max(1e4).refine((ids) => new Set(ids).size === ids.length)
+}).strict();
 const journeyWorkSchema = object({
 	id: journeyIdSchema,
 	title: label,
@@ -10198,6 +10254,7 @@ const journeyWorkSchema = object({
 		/** Images the Work's sessions leave out (D7); absent when a session receipt predates it. */
 		imagesOmitted: count$1.optional()
 	}).strict().optional(),
+	reviewInterruptions: array(reviewInterruptionSchema).optional(),
 	totalBatches: count$1,
 	completedBatches: count$1,
 	groupIds: array(journeyIdSchema),
@@ -10221,7 +10278,12 @@ const journeyWorkSchema = object({
 const journeyConsentSchema = object({
 	provider: label,
 	allowedAt: datetime(),
-	source: _enum(["conversation", "panel"])
+	source: _enum(["conversation", "panel"]),
+	/**
+	* The account signed in when consent was recorded (ADR 0276 D8). Absent in a build with no
+	* sign-in client and in a journey started before this version (Q11 a, Q13 a).
+	*/
+	subject: string().min(1).max(128).optional()
 }).strict();
 const journeyExceptionSchema = object({
 	id: journeyIdSchema,
@@ -10235,6 +10297,7 @@ const journeyExceptionSchema = object({
 }).strict();
 object({
 	formatVersion: literal(10),
+	outputVersion: number().int().positive().optional(),
 	classificationRevision: count$1.nullable(),
 	privacy: journeyPrivacySchema,
 	delegation: journeyDelegationSchema.nullable(),
@@ -10292,7 +10355,21 @@ object({
 	/** Sessions whose last activity lies in this period (ADR 0206); absent means all. */
 	period: sessionPeriodSchema.optional(),
 	/** The Passport address and links chosen in this Journey (ADR 0246); absent means none yet. */
-	profile: journeyProfileSchema.optional()
+	profile: journeyProfileSchema.optional(),
+	/**
+	* The last account read that re-seeded every Work (ADR 0277 section 5). Only scope
+	* confirmation and Retry set `read`; late classification records only a failure. Absent in a
+	* journey of an earlier version, which reads as not `read`.
+	*/
+	accountRead: object({
+		status: _enum([
+			"read",
+			"signed-out",
+			"unavailable"
+		]),
+		subject: string().min(1).max(128).optional(),
+		at: datetime()
+	}).strict().optional()
 }).strict();
 //#endregion
 //#region packages/application/src/journey/b/model.ts
@@ -10616,7 +10693,7 @@ function locateTerms(batches, terms) {
 		}];
 	}));
 	for (const { type, value, index } of unique.values()) {
-		if (admitCustomTerm(value)?.value !== value || !texts.some((item) => item.text.includes(value))) {
+		if (value.includes("█") || admitCustomTerm(value)?.value !== value || !texts.some((item) => item.text.includes(value))) {
 			unlocated++;
 			unlocatedIndices.push(index);
 			continue;
@@ -10640,7 +10717,107 @@ function locateTerms(batches, terms) {
 	};
 }
 //#endregion
+//#region packages/application/src/journey/terms/read-pages.ts
+/** Leave ample space for tool framing, JSON escaping and the final receipt on each read. */
+const pageBytes = 6e3;
+/** Each receipt binds the delivered text, its position, the frozen batch and its assignment. */
+function reviewReadPages(batch, attemptId) {
+	const pages = [];
+	let text = "";
+	function flush() {
+		if (!text) return;
+		const receipt = createHash("sha256").update(JSON.stringify([
+			"scan-read-v1",
+			attemptId,
+			batch.id,
+			batch.digest,
+			pages.length,
+			text
+		])).digest("hex");
+		pages.push({
+			text,
+			receipt
+		});
+		text = "";
+	}
+	for (const chunk of batch.chunks) {
+		if (!chunk.text.length) throw new Error("EMPTY_REVIEW_PAGE");
+		let offset = 0;
+		while (offset < chunk.text.length) {
+			const header = "\nCHUNK " + chunk.id + " " + (chunk.core ? "review" : "context") + " " + chunk.role + " UTF16_START " + offset + "\n";
+			const available = pageBytes - Buffer.byteLength(text + header);
+			if (available < 4) {
+				if (!text) throw new Error("INVALID_REVIEW_PAGE_HEADER");
+				flush();
+				continue;
+			}
+			let end = offset;
+			let bytes = 0;
+			while (end < chunk.text.length) {
+				const point = chunk.text.codePointAt(end);
+				if (point === void 0) throw new Error("INVALID_REVIEW_PAGE_RANGE");
+				const size = point < 128 ? 1 : point < 2048 ? 2 : point < 65536 ? 3 : 4;
+				if (bytes + size > available) break;
+				end += point >= 65536 ? 2 : 1;
+				bytes += size;
+			}
+			text += header + chunk.text.slice(offset, end);
+			offset = end;
+			if (offset < chunk.text.length) flush();
+		}
+	}
+	flush();
+	return pages;
+}
+//#endregion
 //#region apps/cli/src/skill-install/privacy-validator.ts
+const sourceReferenceSchema = object({
+	chunkId: uuid(),
+	start: number().int(),
+	end: number().int()
+}).strict();
+/** Resolve sensitive spans locally against the projected, assigned core text. */
+function resolvePrivacySources(raw, batches) {
+	if (!Array.isArray(raw)) throw new Error("INVALID_FINDING");
+	if (!raw.some((term) => typeof term === "object" && term !== null && "source" in term)) return raw;
+	const chunks = new Map(batches.flatMap(({ chunks }) => chunks.map((chunk) => [chunk.id, chunk])));
+	const diagnostics = [];
+	let invalidCount = 0;
+	const resolved = raw.map((term, index) => {
+		if (typeof term !== "object" || term === null || !("source" in term)) return term;
+		const invalid = (reason) => {
+			invalidCount++;
+			if (diagnostics.length < 32) diagnostics.push(`draft.terms[${index}].source: ${reason}`);
+		};
+		const reference = sourceReferenceSchema.safeParse(term.source);
+		if (!reference.success || "value" in term) return invalid("use only a chunkId and integer start/end offsets for the source");
+		const { chunkId, start, end } = reference.data;
+		const chunk = chunks.get(chunkId);
+		if (!chunk?.core) return invalid("select an assigned core chunk");
+		if (start < 0 || end <= start || end > chunk.text.length) return invalid("check chunk-relative UTF-16 bounds");
+		if (end - start > 8e3) return invalid("span exceeds the length limit");
+		if (chunk.displayMasks?.some((range) => range.start < end && start < range.end)) return invalid("span intersects masked text");
+		const value = chunk.text.slice(start, end);
+		if (!value.isWellFormed()) {
+			if ((chunk.text.codePointAt(start - 1) ?? 0) > 65535 || (chunk.text.codePointAt(end - 1) ?? 0) > 65535) return invalid("offset splits a UTF-16 surrogate pair");
+			return invalid("span contains invalid UTF-16");
+		}
+		if (value.includes("█")) return invalid("span intersects masked text");
+		if (!value.trim()) return invalid("span contains only whitespace");
+		if (Array.from(value).length > 4e3) return invalid("span exceeds the length limit");
+		const { source: _source, ...rest } = term;
+		return {
+			...rest,
+			value
+		};
+	});
+	if (invalidCount) {
+		const remaining = invalidCount - diagnostics.length;
+		if (remaining) diagnostics.push(`${remaining} additional invalid source references`);
+		throw new Error("INVALID_REVIEW_SPAN", { cause: diagnostics });
+	}
+	return resolved;
+}
 const suppliedTerms = array(object({
 	key: string(),
 	type: string(),
@@ -10683,4 +10860,4 @@ function validateStageAnswer(input, answer) {
 	}
 }
 //#endregion
-export { locateTerms, validateStageAnswer };
+export { locateTerms, resolvePrivacySources, reviewReadPages, validateStageAnswer };

package/skills/tacitt/scripts/review.d.mts

@@ -3,11 +3,7 @@ export function parseReply(text: string): Record<string, unknown>
 export function estimateTokens(text: string): number
 export function readPlan(text: string, pageTokens?: number): { offset: number; limit: number }[]
 export function renderGuidance(guidance: unknown): string
-export function renderPackTask(
-  packId: string,
-  guidance: unknown,
-  units: readonly { ref: string; unit: unknown }[],
-): string
+export function renderPackTask(packId: string, guidance: unknown): string
 export function readAnswer(replies: readonly Record<string, unknown>[]): {
   units: Map<string, { summary: string; exclude: string | null }>
   terms: { type: string; value: string; disposition: string; reason: string }[]
@@ -74,9 +70,13 @@ export function writePackTask(
   task: string,
   text: string,
   unitRefs: readonly string[],
-  sourceBatches?: readonly {
+  sourceBatches: readonly {
+    id: string
+    digest: string
     workId: string
     chunks: readonly {
+      id: string
+      role: string
       text: string
       offset: number
       core: boolean
@@ -84,4 +84,5 @@ export function writePackTask(
       fieldHash: string
     }[]
   }[],
+  attemptId?: string,
 ): Promise<string>