Add GPT-based code review

.github/workflows/code-review.yaml

@@ -0,0 +1,29 @@
+name: GPT Code Review 🤖
+
+permissions:
+  contents: read
+  pull-requests: write
+
+on:
+  pull_request:
+    types:
+      - opened
+      - reopened
+      - synchronize
+
+jobs:
+  review:
+    name: GPT Code Review 🤖
+    runs-on: ubuntu-latest
+    steps:
+      - name: Checkout Repo 🛎️
+        uses: actions/checkout@v3
+        with:
+          fetch-depth: 0
+
+      - name: GPT Code Review 🤖
+        uses: mattzcarey/[email protected]
+        with:
+          OPENAI_API_KEY: ${{ secrets.OPENAI_API_KEY }}
+          MODEL: 'gpt-3.5-turbo'
+          GITHUB_TOKEN: ${{ github.token }}

randomfile.ts

@@ -0,0 +1,97 @@
+/* eslint-disable max-depth */
+/* eslint-disable complexity */
+// Reviewing multiple files inline > prioritising them > adding review comments
+// Answer questions > get the comments on the PR (by me and the questioner) as context > answer the question as comment
+
+import jsesc from "jsesc";
+
+import { modelInfo } from "../constants";
+import { AIModel } from "../llm/ai";
+import { buildReviewPrompt } from "../prompts/buildPrompt";
+import { ReviewFile } from "../types";
+
+export class Chat {
+  ai: AIModel;
+  modelName: string;
+  constructor(
+    openaiApiKey: string,
+    openaiModelName?: string,
+    temperature?: string
+  ) {
+    this.modelName = openaiModelName ?? "gpt-4-1106-preview";
+    this.ai = new AIModel({
+      modelName: this.modelName,
+      apiKey: openaiApiKey,
+      temperature: temperature ? parseFloat(temperature) : 0,
+    });
+  }
+
+  private getMaxPromptLength = (modelName: string): number => {
+    const model = modelInfo.find((info) => info.model === modelName);
+    if (!model) {
+      throw new Error(`Model ${modelName} not found`);
+    }
+
+    return model.maxPromptLength;
+  };
+
+  public getReview = async (
+    patch: string
+  ): Promise<ReviewFile[] | undefined> => {
+    const prompt = buildReviewPrompt(patch);
+    const maxPromptLength = this.getMaxPromptLength(this.modelName);
+
+    if (prompt.length > maxPromptLength) {
+      console.error(
+        `File ${prompt} is too large to review, skipping review for this file`
+      );
+
+      return undefined;
+    }
+
+    try {
+      let jsonResponse = await this.ai.callModel(prompt);
+      jsonResponse = removeMarkdownJsonQuotes(jsonResponse);
+
+      try {
+        return JSON.parse(jsonResponse) as ReviewFile[];
+      } catch (parseError) {
+        console.error(
+          `Error parsing JSON: ${
+            (parseError as Error).message
+          }. Escaping special characters and retrying.`
+        );
+
+        try {
+          // eslint-disable-next-line @typescript-eslint/no-unsafe-assignment, @typescript-eslint/no-unsafe-call
+          const escapedJsonResponse: string = jsesc(jsonResponse, {
+            json: true,
+          });
+
+          return JSON.parse(escapedJsonResponse) as ReviewFile[];
+        } catch (escapeParseError) {
+          console.error(
+            `Error parsing escaped JSON: ${
+              (escapeParseError as Error).message
+            }. Returning undefined.`
+          );
+
+          return undefined;
+        }
+      }
+    } catch (error) {
+      console.error(
+        `Error processing review data: ${(error as Error).message}`
+      );
+
+      return undefined;
+    }
+  };
+}
+
+const removeMarkdownJsonQuotes = (jsonString: string): string => {
+  return jsonString
+    .replace(/^`+\s*json\s*/, "")
+    .replace(/\s*`+$/, "")
+    .trim();
+};