@aaronshaf/ger 0.1.5 → 0.1.6

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@aaronshaf/ger",
-  "version": "0.1.5",
+  "version": "0.1.6",
   "module": "index.ts",
   "type": "module",
   "bin": {

package/src/cli/commands/review.ts

@@ -89,7 +89,7 @@ interface ReviewOptions {
   comment?: boolean
   yes?: boolean
   prompt?: string
-  provider?: string
+  tool?: string
   systemPrompt?: string
 }
 
@@ -336,7 +336,7 @@ export const reviewCommand = (changeId: string, options: ReviewOptions = {}) =>
     }
 
     // Select strategy based on user preference
-    const selectedStrategy = yield* reviewStrategy.selectStrategy(options.provider)
+    const selectedStrategy = yield* reviewStrategy.selectStrategy(options.tool)
     yield* Console.log(`✓ Using AI tool: ${selectedStrategy.name}`)
 
     // Load custom review prompt if provided

package/src/cli/index.ts

@@ -368,12 +368,12 @@ program
   .option('-y, --yes', 'Skip confirmation prompts when posting comments')
   .option('--debug', 'Show debug output including AI responses')
   .option('--prompt <file>', 'Path to custom review prompt file (e.g., ~/prompts/review.md)')
-  .option('--provider <provider>', 'Preferred AI provider (claude-sdk, claude, gemini, opencode)')
+  .option('--tool <tool>', 'Preferred AI tool (claude, gemini, opencode)')
   .option('--system-prompt <prompt>', 'Custom system prompt for the AI')
   .addHelpText(
     'after',
     `
-This command uses AI (Claude SDK, claude CLI, gemini CLI, or opencode CLI) to review a Gerrit change.
+This command uses AI (claude CLI, gemini CLI, or opencode CLI) to review a Gerrit change.
 It performs a two-stage review process:
 
 1. Generates inline comments for specific code issues
@@ -384,7 +384,7 @@ Use --comment to post the review to Gerrit (with confirmation prompts).
 Use --comment --yes to post without confirmation.
 
 Requirements:
-  - One of these AI tools must be available: Claude SDK (ANTHROPIC_API_KEY), claude CLI, gemini CLI, or opencode CLI
+  - One of these AI tools must be available: claude CLI, gemini CLI, or opencode CLI
   - Gerrit credentials must be configured (run 'ger setup' first)
 
 Examples:
@@ -397,6 +397,9 @@ Examples:
   # Review and auto-post comments without prompting
   $ ger review 12345 --comment --yes
   
+  # Use specific AI tool
+  $ ger review 12345 --tool gemini
+  
   # Show debug output to troubleshoot issues
   $ ger review 12345 --debug
 `,
@@ -408,7 +411,7 @@ Examples:
         yes: options.yes,
         debug: options.debug,
         prompt: options.prompt,
-        provider: options.provider,
+        tool: options.tool,
         systemPrompt: options.systemPrompt,
       }).pipe(
         Effect.provide(ReviewStrategyServiceLive),

package/src/prompts/system-inline-review.md

@@ -1,9 +1,15 @@
-## CRITICAL OUTPUT REQUIREMENT
+## IMMEDIATE TASK: ANALYZE CODE AND GENERATE INLINE COMMENTS
 
-**YOUR ENTIRE OUTPUT MUST BE WRAPPED IN <response></response> TAGS.**
-**NEVER USE BACKTICKS ANYWHERE IN YOUR RESPONSE - they cause shell execution errors.**
+**YOU MUST ANALYZE THE PROVIDED CODE CHANGES RIGHT NOW AND GENERATE INLINE COMMENTS.**
 
-Output ONLY a JSON array wrapped in response tags. **EMPTY ARRAY IS PERFECTLY VALID** for clean code without issues. No other text before or after the tags.
+**CRITICAL OUTPUT REQUIREMENT:**
+- YOUR ENTIRE OUTPUT MUST BE WRAPPED IN <response></response> TAGS
+- NEVER USE BACKTICKS ANYWHERE IN YOUR RESPONSE - they cause shell execution errors  
+- Output ONLY a JSON array wrapped in response tags
+- **EMPTY ARRAY IS PERFECTLY VALID** for clean code without issues
+- No other text before or after the tags
+
+**START YOUR ANALYSIS NOW. DO NOT ASK QUESTIONS. DO NOT WAIT FOR MORE INPUT.**
 
 ## JSON Structure for Inline Comments
 
@@ -21,12 +27,17 @@ The JSON array must contain inline comment objects with these fields:
   - "start_character": Optional column start (integer)
   - "end_character": Optional column end (integer)
 
-**IMPORTANT**: Every comment MUST have either "line" OR "range". Comments without valid line numbers will be rejected.
+**CRITICAL LINE NUMBER RULES:**
+1. **ALWAYS use final file line numbers, NEVER diff line numbers**
+2. Line numbers must match the NEW version of the file after all changes
+3. Use `git show HEAD:path/to/file` or examine the final file to get correct line numbers
+4. If you see "+50" in a diff, the actual line number is NOT 50 - check the final file
+5. Every comment MUST have either "line" OR "range". Comments without valid line numbers will be rejected.
 
 ### Optional Fields
 - "side": "REVISION" (new code, default) or "PARENT" (original code)
 
-Line numbers refer to the final file (REVISION), not the diff.
+**VERIFICATION STEP**: Before adding any comment, verify the line number by checking the final file content to ensure your line number points to the exact code you're commenting on.
 
 ## Comment Quality Guidelines
 
@@ -85,11 +96,11 @@ You are running in a git repository with full access to:
 - All project files for architectural understanding
 - Use these commands to provide comprehensive, accurate reviews
 
-## FINAL REMINDER
+## FINAL TASK INSTRUCTION
 
-**CRITICAL: Your ENTIRE output must be a JSON array wrapped in <response></response> tags.**
+**ANALYZE THE CODE CHANGES NOW AND OUTPUT YOUR INLINE COMMENTS IMMEDIATELY.**
 
-Example formats:
+Your output format must be:
 ```
 <response>
 []
@@ -97,6 +108,8 @@ Example formats:
 ```
 (Empty array for clean code - this is GOOD!)
 
+OR:
+
 ```
 <response>
 [{"file": "auth.js", "line": 42, "message": "🤖 SQL injection vulnerability: query uses string concatenation"}]
@@ -104,10 +117,12 @@ Example formats:
 ```
 (Only comment on real problems)
 
-**REQUIREMENTS**:
+**CRITICAL REQUIREMENTS**:
 - Every message must start with "🤖 "
 - Never use backticks in your response
 - Empty arrays are encouraged for clean code
 - Focus on bugs, security, crashes - ignore style preferences  
 - Use git commands to understand context before commenting
-- NO TEXT OUTSIDE THE <response></response> TAGS
+- NO TEXT OUTSIDE THE <response></response> TAGS
+
+**DO YOUR ANALYSIS NOW. STOP ASKING QUESTIONS. GENERATE THE REVIEW.**

package/src/prompts/system-overall-review.md

@@ -41,11 +41,16 @@ Gerrit uses a LIMITED markdown subset. Follow these rules EXACTLY:
 
 **YOUR ENTIRE OUTPUT MUST BE WRAPPED IN <response></response> TAGS.**
 
+**IMMEDIATE TASK**: Analyze the code changes provided below and write a comprehensive engineering review.
+
 Start with "🤖 [Your Tool Name] ([Your Model])" then provide a **CONCISE** engineering assessment. Examples:
 - If you are Claude Sonnet 4: "🤖 Claude (Sonnet 4)"
+- If you are Gemini: "🤖 Gemini (1.5 Pro)" or "🤖 Gemini (1.5 Flash)"
 - For clean code: "No significant issues found. Change is ready for merge."
 - For problematic code: Focus only on critical/important issues, skip minor concerns
 
+**YOU MUST ANALYZE THE PROVIDED CODE CHANGES AND WRITE A REVIEW NOW.**
+
 ## Example Output Format
 
 <response>
@@ -175,3 +180,9 @@ CRITICAL FORMATTING RULES:
 - Use exactly 4 spaces to start each line of code blocks  
 - Keep code blocks simple and readable
 - Add proper spacing for readability
+
+## TASK SUMMARY
+
+**ANALYZE THE CODE CHANGES PROVIDED ABOVE AND WRITE YOUR ENGINEERING REVIEW IMMEDIATELY.**
+
+Do not ask for clarification. Do not wait for more input. Analyze the provided code changes, git history, and changed files, then write your review following the format requirements above.

package/src/services/review-strategy.ts

@@ -4,6 +4,13 @@ import { promisify } from 'node:util'
 
 const execAsync = promisify(exec)
 
+// Shared response extraction logic for all AI tools
+const extractResponse = (stdout: string): string => {
+  // Extract response from <response> tags or use full output
+  const responseMatch = stdout.match(/<response>([\s\S]*?)<\/response>/i)
+  return responseMatch ? responseMatch[1].trim() : stdout.trim()
+}
+
 // Simple strategy focused only on review needs
 export class ReviewStrategyError extends Data.TaggedError('ReviewStrategyError')<{
   message: string
@@ -75,9 +82,7 @@ export const claudeCliStrategy: ReviewStrategy = {
           }),
       })
 
-      // Extract response from <response> tags or use full output
-      const responseMatch = result.stdout.match(/<response>([\s\S]*?)<\/response>/i)
-      return responseMatch ? responseMatch[1].trim() : result.stdout.trim()
+      return extractResponse(result.stdout)
     }),
 }
 
@@ -135,7 +140,7 @@ export const geminiCliStrategy: ReviewStrategy = {
           }),
       })
 
-      return result.stdout.trim()
+      return extractResponse(result.stdout)
     }),
 }
 
@@ -193,7 +198,7 @@ export const openCodeCliStrategy: ReviewStrategy = {
           }),
       })
 
-      return result.stdout.trim()
+      return extractResponse(result.stdout)
     }),
 }
 

package/tests/unit/services/review-strategy.test.ts

@@ -7,11 +7,6 @@ interface MockDeps {
   spawn: (command: string, options: any) => any
 }
 
-// Mock for Claude SDK
-interface MockClaudeSDK {
-  query: any // Will be a Bun mock function
-}
-
 // Test implementation that mirrors the real strategy structure
 const createTestStrategy = (name: string, command: string, flags: string[], deps: MockDeps) => ({
   name,
@@ -74,47 +69,6 @@ const createTestStrategy = (name: string, command: string, flags: string[], deps
     }),
 })
 
-// Test implementation for Claude SDK strategy
-const createSDKStrategy = (
-  name: string,
-  deps: { sdk: MockClaudeSDK | null; hasApiKey: boolean },
-) => ({
-  name,
-  isAvailable: () =>
-    Effect.gen(function* () {
-      return Boolean(deps.sdk && deps.hasApiKey)
-    }),
-  executeReview: (prompt: string, options: { cwd?: string; systemPrompt?: string } = {}) =>
-    Effect.gen(function* () {
-      if (!deps.sdk) {
-        return yield* Effect.fail(new Error(`${name} not available`))
-      }
-
-      const result = yield* Effect.tryPromise({
-        try: async () => {
-          for await (const message of deps.sdk!.query({
-            prompt,
-            options: {
-              maxTurns: 3,
-              customSystemPrompt: options.systemPrompt || 'You are a code review expert.',
-              allowedTools: ['Read', 'Grep', 'Glob'],
-              cwd: options.cwd,
-            },
-          })) {
-            if (message.type === 'result' && message.subtype === 'success' && message.result) {
-              return message.result
-            }
-          }
-          throw new Error('No result received')
-        },
-        catch: (error) =>
-          new Error(`${name} failed: ${error instanceof Error ? error.message : String(error)}`),
-      })
-
-      return result
-    }),
-})
-
 describe('Review Strategy', () => {
   let mockExecAsync: any
   let mockSpawn: any
@@ -266,6 +220,14 @@ describe('Review Strategy', () => {
       expect(response).toBe('Gemini response')
       expect(mockSpawn).toHaveBeenCalledWith('gemini -p', expect.any(Object))
     })
+
+    it('should extract response from tags', async () => {
+      setupSuccessfulExecution('<response>Gemini tagged content</response>')
+
+      const response = await Effect.runPromise(geminiStrategy.executeReview('Test prompt'))
+
+      expect(response).toBe('Gemini tagged content')
+    })
   })
 
   describe('OpenCode CLI Strategy', () => {
@@ -295,120 +257,13 @@ describe('Review Strategy', () => {
       expect(response).toBe('OpenCode response')
       expect(mockSpawn).toHaveBeenCalledWith('opencode -p', expect.any(Object))
     })
-  })
 
-  describe('Claude SDK Strategy', () => {
-    let mockSDK: MockClaudeSDK
-    let claudeSDKStrategy: any
-
-    beforeEach(() => {
-      mockSDK = {
-        query: mock(),
-      }
-
-      claudeSDKStrategy = createSDKStrategy('Claude SDK', {
-        sdk: mockSDK,
-        hasApiKey: true,
-      })
-    })
-
-    it('should check availability when SDK and API key are present', async () => {
-      const available = await Effect.runPromise(claudeSDKStrategy.isAvailable())
-
-      expect(available).toBe(true)
-    })
-
-    it('should check availability when SDK is missing', async () => {
-      const strategy = createSDKStrategy('Claude SDK', {
-        sdk: null,
-        hasApiKey: true,
-      })
-
-      const available = await Effect.runPromise(strategy.isAvailable())
-
-      expect(available).toBe(false)
-    })
-
-    it('should check availability when API key is missing', async () => {
-      const strategy = createSDKStrategy('Claude SDK', {
-        sdk: mockSDK,
-        hasApiKey: false,
-      })
-
-      const available = await Effect.runPromise(strategy.isAvailable())
-
-      expect(available).toBe(false)
-    })
-
-    it('should execute review successfully', async () => {
-      mockSDK.query = mock(async function* () {
-        yield { type: 'message', content: 'Thinking...' }
-        yield { type: 'result', subtype: 'success', result: 'Claude SDK response' }
-      })
-
-      const response = await Effect.runPromise(
-        claudeSDKStrategy.executeReview('Test prompt', { cwd: '/tmp' }),
-      )
-
-      expect(response).toBe('Claude SDK response')
-      expect(mockSDK.query).toHaveBeenCalledWith({
-        prompt: 'Test prompt',
-        options: {
-          maxTurns: 3,
-          customSystemPrompt: 'You are a code review expert.',
-          allowedTools: ['Read', 'Grep', 'Glob'],
-          cwd: '/tmp',
-        },
-      })
-    })
-
-    it('should use custom system prompt', async () => {
-      mockSDK.query = mock(async function* () {
-        yield { type: 'result', subtype: 'success', result: 'Custom prompt response' }
-      })
-
-      const response = await Effect.runPromise(
-        claudeSDKStrategy.executeReview('Test prompt', {
-          systemPrompt: 'Custom review prompt',
-        }),
-      )
-
-      expect(response).toBe('Custom prompt response')
-      expect(mockSDK.query).toHaveBeenCalledWith({
-        prompt: 'Test prompt',
-        options: {
-          maxTurns: 3,
-          customSystemPrompt: 'Custom review prompt',
-          allowedTools: ['Read', 'Grep', 'Glob'],
-          cwd: undefined,
-        },
-      })
-    })
-
-    it('should handle SDK failure', async () => {
-      mockSDK.query = mock(async function* () {
-        throw new Error('SDK error')
-      })
-
-      try {
-        await Effect.runPromise(claudeSDKStrategy.executeReview('Test prompt'))
-        expect(false).toBe(true) // Should not reach here
-      } catch (error: any) {
-        expect(error.message).toContain('Claude SDK failed')
-      }
-    })
+    it('should extract response from tags', async () => {
+      setupSuccessfulExecution('<response>OpenCode tagged content</response>')
 
-    it('should handle no result received', async () => {
-      mockSDK.query = mock(async function* () {
-        yield { type: 'message', content: 'No result' }
-      })
+      const response = await Effect.runPromise(opencodeStrategy.executeReview('Test prompt'))
 
-      try {
-        await Effect.runPromise(claudeSDKStrategy.executeReview('Test prompt'))
-        expect(false).toBe(true) // Should not reach here
-      } catch (error: any) {
-        expect(error.message).toContain('No result received')
-      }
+      expect(response).toBe('OpenCode tagged content')
     })
   })