@unifan/pi-review-zh 1.0.0

package/src/workflow-schemas.ts

@@ -0,0 +1,172 @@
+/**
+ * JSON Schema fragments injected as `outputSchema` on each reviewer and the
+ * gate. Kept in sync with src/schema.ts (used by the legacy spawn pipeline).
+ *
+ * The plugin no longer reads reviewer Markdown text; the workflowScript
+ * receives validated `result.structuredOutput` from pi-subagents.
+ */
+import type { IssueSeverity, IssueCategory, Verdict } from "./types.js";
+
+export type JsonSchema = Record<string, unknown>;
+
+const SEVERITY: IssueSeverity[] = ["blocker", "major", "minor", "nit"];
+const CATEGORY: IssueCategory[] = [
+	"compliance",
+	"bug",
+	"convention",
+	"history",
+	"security",
+	"performance",
+	"docs",
+	"other",
+];
+const VERDICT: Verdict[] = ["approve", "request_changes", "comment"];
+
+/**
+ * Stable JSON Schema describing the structured output of every reviewer.
+ * `status` distinguishes runtime `ok` (tool succeeded) from business
+ * `limited/skipped` (no findings because of missing context, not silence).
+ */
+export const REVIEWER_OUTPUT_SCHEMA: JsonSchema = {
+	type: "object",
+	additionalProperties: false,
+	required: ["status", "issues", "summary", "coverage"],
+	properties: {
+		status: {
+			type: "string",
+			enum: ["ok", "limited", "skipped"],
+			description:
+				"ok = reviewer produced findings normally; limited = could not run some checks (still may have issues); skipped = lane does not apply.",
+		},
+		issues: {
+			type: "array",
+			maxItems: 200,
+			items: {
+				type: "object",
+				additionalProperties: false,
+				required: ["file", "category", "severity", "confidence", "evidence", "fingerprint"],
+				properties: {
+					file: { type: "string", minLength: 1, maxLength: 500 },
+					line: { type: "integer", minimum: 1 },
+					relatedChangedLine: { type: "integer", minimum: 1 },
+					category: { type: "string", enum: CATEGORY },
+					severity: { type: "string", enum: SEVERITY },
+					confidence: { type: "integer", minimum: 1, maximum: 10 },
+					evidence: { type: "string", minLength: 1, maxLength: 280 },
+					fingerprint: {
+						type: "string",
+						minLength: 1,
+						maxLength: 80,
+						description:
+							"Stable id (file:line:category[:short-hash-of-evidence]) so the gate can dedupe across reviewers.",
+					},
+				},
+			},
+		},
+		summary: { type: "string", minLength: 1, maxLength: 2000 },
+		coverage: {
+			type: "object",
+			additionalProperties: false,
+			required: ["filesChecked", "commandsRun", "limitations"],
+			properties: {
+				filesChecked: { type: "array", items: { type: "string" } },
+				commandsRun: { type: "array", items: { type: "string" } },
+				limitations: {
+					type: "array",
+					items: { type: "string" },
+					description: "Plain-language reasons for limited/skipped status.",
+				},
+			},
+		},
+	},
+};
+
+/**
+ * Stable JSON Schema describing the gate's structured output. Adds
+ * `dispositions` so every candidate (kept/dropped/merged) is auditable.
+ */
+export const GATE_OUTPUT_SCHEMA: JsonSchema = {
+	type: "object",
+	additionalProperties: false,
+	required: ["status", "verdict", "issues", "dispositions", "reason"],
+	properties: {
+		status: {
+			type: "string",
+			enum: ["ok", "limited", "skipped"],
+		},
+		verdict: { type: "string", enum: VERDICT },
+		issues: {
+			type: "array",
+			maxItems: 500,
+			items: {
+				type: "object",
+				additionalProperties: false,
+				required: ["file", "category", "severity", "confidence", "evidence", "fingerprint"],
+				properties: {
+					file: { type: "string", minLength: 1, maxLength: 500 },
+					line: { type: "integer", minimum: 1 },
+					relatedChangedLine: { type: "integer", minimum: 1 },
+					category: { type: "string", enum: CATEGORY },
+					severity: { type: "string", enum: SEVERITY },
+					confidence: { type: "integer", minimum: 1, maximum: 10 },
+					evidence: { type: "string", minLength: 1, maxLength: 280 },
+					fingerprint: { type: "string", minLength: 1, maxLength: 80 },
+				},
+			},
+		},
+		dispositions: {
+			type: "array",
+			description:
+				"Per-candidate disposition: kept | dropped | merged. Required for every reviewer candidate so reports can audit threshold/dedupe/verify decisions.",
+			items: {
+				type: "object",
+				additionalProperties: false,
+				required: ["fingerprint", "decision", "originalConfidence", "finalConfidence", "reason"],
+				properties: {
+					fingerprint: { type: "string", minLength: 1, maxLength: 80 },
+					decision: { type: "string", enum: ["kept", "dropped", "merged"] },
+					originalConfidence: { type: "integer", minimum: 1, maximum: 10 },
+					finalConfidence: { type: "integer", minimum: 1, maximum: 10 },
+					sourceReviewers: {
+						type: "array",
+						items: { type: "string" },
+					},
+					reason: { type: "string", minLength: 1, maxLength: 500 },
+				},
+			},
+		},
+		reason: { type: "string", minLength: 1, maxLength: 500 },
+		coverage: {
+			type: "object",
+			additionalProperties: false,
+			properties: {
+				limitations: { type: "array", items: { type: "string" } },
+			},
+		},
+	},
+};
+
+/**
+ * Serialize a JSON Schema as a JS string literal safe to embed in the
+ * generated workflowScript. Avoids escaping mistakes from JSON.stringify +
+ * double-quoting gymnastics.
+ */
+export function serializeSchemaForJs(schema: JsonSchema): string {
+	return JSON.stringify(schema);
+}
+
+/**
+ * Serialize a JSON Schema as a multi-line JS object literal for embedding
+ * directly in the generated workflowScript (v0.7.3). The single-line form
+ * produced 1400+ character lines — the exact spot where the main agent's
+ * copy slipped a character (PR 19291 incident, `"maxLength":80",`).
+ * Multi-line keeps every line short enough to copy reliably.
+ */
+export function serializeSchemaAsObjectLiteral(schema: JsonSchema, indent = "  "): string {
+	const json = JSON.stringify(schema, null, 2);
+	// Indent every line so the literal sits cleanly inside the script body.
+	return json
+		.split("\n")
+		.map((line, i) => (i === 0 ? line : indent + line))
+		.join("\n");
+}