refactor(ai): Implement unified AI service layer and fix subtask update
- Unified Service: Introduced 'scripts/modules/ai-services-unified.js' to centralize AI interactions using provider modules ('src/ai-providers/') and the Vercel AI SDK.
- Provider Modules: Implemented 'anthropic.js' and 'perplexity.js' wrappers for Vercel SDK.
- 'updateSubtaskById' Fix: Refactored the AI call within 'updateSubtaskById' to use 'generateTextService' from the unified layer, resolving runtime errors related to parameter passing and streaming. This serves as the pattern for refactoring other AI calls in 'scripts/modules/task-manager/'.
- Task Status: Marked Subtask 61.19 as 'done'.
- Rules: Added new 'ai-services.mdc' rule.
This centralizes AI logic, replacing previous direct SDK calls and custom implementations. API keys are resolved via 'resolveEnvVariable' within the service layer. The refactoring of 'updateSubtaskById' establishes the standard approach for migrating other AI-dependent functions in the task manager module to use the unified service.
Relates to Task 61.
This commit is contained in:
Eyal Toledano
@@ -10,7 +10,7 @@ import {
|
||||
stopLoadingIndicator
|
||||
} from '../ui.js';
|
||||
import { log, readJSON, writeJSON, truncate, isSilentMode } from '../utils.js';
|
||||
import { getAvailableAIModel } from '../ai-services.js';
|
||||
import { generateTextService } from '../ai-services-unified.js';
|
||||
import {
|
||||
getDebugFlag,
|
||||
getMainModelId,
|
||||
@@ -54,6 +54,7 @@ async function updateSubtaskById(
|
||||
};
|
||||
|
||||
let loadingIndicator = null;
|
||||
|
||||
try {
|
||||
report(`Updating subtask ${subtaskId} with prompt: "${prompt}"`, 'info');
|
||||
|
||||
@@ -193,236 +194,55 @@ async function updateSubtaskById(
|
||||
);
|
||||
}
|
||||
|
||||
// Create the system prompt (as before)
|
||||
const systemPrompt = `You are an AI assistant helping to update software development subtasks with additional information.
|
||||
let additionalInformation = '';
|
||||
try {
|
||||
// Reverted: Keep the original system prompt
|
||||
const systemPrompt = `You are an AI assistant helping to update software development subtasks with additional information.
|
||||
Given a subtask, you will provide additional details, implementation notes, or technical insights based on user request.
|
||||
Focus only on adding content that enhances the subtask - don't repeat existing information.
|
||||
Be technical, specific, and implementation-focused rather than general.
|
||||
Provide concrete examples, code snippets, or implementation details when relevant.`;
|
||||
|
||||
// Replace the old research/Claude code with the new model selection approach
|
||||
let additionalInformation = '';
|
||||
let modelAttempts = 0;
|
||||
const maxModelAttempts = 2; // Try up to 2 models before giving up
|
||||
// Reverted: Use the full JSON stringification for the user message
|
||||
const subtaskData = JSON.stringify(subtask, null, 2);
|
||||
const userMessageContent = `Here is the subtask to enhance:\n${subtaskData}\n\nPlease provide additional information addressing this request:\n${prompt}\n\nReturn ONLY the new information to add - do not repeat existing content.`;
|
||||
|
||||
while (modelAttempts < maxModelAttempts && !additionalInformation) {
|
||||
modelAttempts++; // Increment attempt counter at the start
|
||||
const isLastAttempt = modelAttempts >= maxModelAttempts;
|
||||
let modelType = null; // Declare modelType outside the try block
|
||||
const serviceRole = useResearch ? 'research' : 'main';
|
||||
report(`Calling AI stream service with role: ${serviceRole}`, 'info');
|
||||
|
||||
try {
|
||||
// Get the best available model based on our current state
|
||||
const result = getAvailableAIModel({
|
||||
claudeOverloaded,
|
||||
requiresResearch: useResearch
|
||||
});
|
||||
modelType = result.type;
|
||||
const client = result.client;
|
||||
const streamResult = await generateTextService({
|
||||
role: serviceRole,
|
||||
session: session,
|
||||
systemPrompt: systemPrompt, // Pass the original system prompt
|
||||
prompt: userMessageContent // Pass the original user message content
|
||||
});
|
||||
|
||||
report(
|
||||
`Attempt ${modelAttempts}/${maxModelAttempts}: Generating subtask info using ${modelType}`,
|
||||
'info'
|
||||
);
|
||||
|
||||
// Update loading indicator text - only for text output
|
||||
if (outputFormat === 'text') {
|
||||
if (loadingIndicator) {
|
||||
stopLoadingIndicator(loadingIndicator); // Stop previous indicator
|
||||
}
|
||||
loadingIndicator = startLoadingIndicator(
|
||||
`Attempt ${modelAttempts}: Using ${modelType.toUpperCase()}...`
|
||||
);
|
||||
}
|
||||
|
||||
const subtaskData = JSON.stringify(subtask, null, 2);
|
||||
const userMessageContent = `Here is the subtask to enhance:\n${subtaskData}\n\nPlease provide additional information addressing this request:\n${prompt}\n\nReturn ONLY the new information to add - do not repeat existing content.`;
|
||||
|
||||
if (modelType === 'perplexity') {
|
||||
// Construct Perplexity payload
|
||||
const perplexityModel = getResearchModelId(session);
|
||||
const response = await client.chat.completions.create({
|
||||
model: perplexityModel,
|
||||
messages: [
|
||||
{ role: 'system', content: systemPrompt },
|
||||
{ role: 'user', content: userMessageContent }
|
||||
],
|
||||
temperature: getResearchTemperature(session),
|
||||
max_tokens: getResearchMaxTokens(session)
|
||||
});
|
||||
additionalInformation = response.choices[0].message.content.trim();
|
||||
} else {
|
||||
// Claude
|
||||
let responseText = '';
|
||||
let streamingInterval = null;
|
||||
|
||||
try {
|
||||
// Only update streaming indicator for text output
|
||||
if (outputFormat === 'text') {
|
||||
let dotCount = 0;
|
||||
const readline = await import('readline');
|
||||
streamingInterval = setInterval(() => {
|
||||
readline.cursorTo(process.stdout, 0);
|
||||
process.stdout.write(
|
||||
`Receiving streaming response from Claude${'.'.repeat(dotCount)}`
|
||||
);
|
||||
dotCount = (dotCount + 1) % 4;
|
||||
}, 500);
|
||||
}
|
||||
|
||||
// Construct Claude payload using config getters
|
||||
const stream = await client.messages.create({
|
||||
model: getMainModelId(session),
|
||||
max_tokens: getMainMaxTokens(session),
|
||||
temperature: getMainTemperature(session),
|
||||
system: systemPrompt,
|
||||
messages: [{ role: 'user', content: userMessageContent }],
|
||||
stream: true
|
||||
});
|
||||
|
||||
for await (const chunk of stream) {
|
||||
if (chunk.type === 'content_block_delta' && chunk.delta.text) {
|
||||
responseText += chunk.delta.text;
|
||||
}
|
||||
if (reportProgress) {
|
||||
await reportProgress({
|
||||
progress:
|
||||
(responseText.length / getMainMaxTokens(session)) * 100
|
||||
});
|
||||
}
|
||||
if (mcpLog) {
|
||||
mcpLog.info(
|
||||
`Progress: ${(responseText.length / getMainMaxTokens(session)) * 100}%`
|
||||
);
|
||||
}
|
||||
}
|
||||
} finally {
|
||||
if (streamingInterval) clearInterval(streamingInterval);
|
||||
// Clear the loading dots line - only for text output
|
||||
if (outputFormat === 'text') {
|
||||
const readline = await import('readline');
|
||||
readline.cursorTo(process.stdout, 0);
|
||||
process.stdout.clearLine(0);
|
||||
}
|
||||
}
|
||||
|
||||
report(
|
||||
`Completed streaming response from Claude API! (Attempt ${modelAttempts})`,
|
||||
'info'
|
||||
);
|
||||
additionalInformation = responseText.trim();
|
||||
}
|
||||
|
||||
// Success - break the loop
|
||||
if (additionalInformation) {
|
||||
report(
|
||||
`Successfully generated information using ${modelType} on attempt ${modelAttempts}.`,
|
||||
'info'
|
||||
);
|
||||
break;
|
||||
} else {
|
||||
// Handle case where AI gave empty response without erroring
|
||||
report(
|
||||
`AI (${modelType}) returned empty response on attempt ${modelAttempts}.`,
|
||||
'warn'
|
||||
);
|
||||
if (isLastAttempt) {
|
||||
throw new Error(
|
||||
'AI returned empty response after maximum attempts.'
|
||||
);
|
||||
}
|
||||
// Allow loop to continue to try another model/attempt if possible
|
||||
}
|
||||
} catch (modelError) {
|
||||
const failedModel =
|
||||
modelType || modelError.modelType || 'unknown model';
|
||||
report(
|
||||
`Attempt ${modelAttempts} failed using ${failedModel}: ${modelError.message}`,
|
||||
'warn'
|
||||
);
|
||||
|
||||
// --- More robust overload check ---
|
||||
let isOverload = false;
|
||||
// Check 1: SDK specific property (common pattern)
|
||||
if (modelError.type === 'overloaded_error') {
|
||||
isOverload = true;
|
||||
}
|
||||
// Check 2: Check nested error property (as originally intended)
|
||||
else if (modelError.error?.type === 'overloaded_error') {
|
||||
isOverload = true;
|
||||
}
|
||||
// Check 3: Check status code if available (e.g., 429 Too Many Requests or 529 Overloaded)
|
||||
else if (modelError.status === 429 || modelError.status === 529) {
|
||||
isOverload = true;
|
||||
}
|
||||
// Check 4: Check the message string itself (less reliable)
|
||||
else if (modelError.message?.toLowerCase().includes('overloaded')) {
|
||||
isOverload = true;
|
||||
}
|
||||
// --- End robust check ---
|
||||
|
||||
if (isOverload) {
|
||||
// Use the result of the check
|
||||
claudeOverloaded = true; // Mark Claude as overloaded for the *next* potential attempt
|
||||
if (!isLastAttempt) {
|
||||
report(
|
||||
'Claude overloaded. Will attempt fallback model if available.',
|
||||
'info'
|
||||
);
|
||||
// Stop the current indicator before continuing - only for text output
|
||||
if (outputFormat === 'text' && loadingIndicator) {
|
||||
stopLoadingIndicator(loadingIndicator);
|
||||
loadingIndicator = null; // Reset indicator
|
||||
}
|
||||
continue; // Go to next iteration of the while loop to try fallback
|
||||
} else {
|
||||
// It was the last attempt, and it failed due to overload
|
||||
report(
|
||||
`Overload error on final attempt (${modelAttempts}/${maxModelAttempts}). No fallback possible.`,
|
||||
'error'
|
||||
);
|
||||
// Let the error be thrown after the loop finishes, as additionalInformation will be empty.
|
||||
// We don't throw immediately here, let the loop exit and the check after the loop handle it.
|
||||
}
|
||||
} else {
|
||||
// Error was NOT an overload
|
||||
// If it's not an overload, throw it immediately to be caught by the outer catch.
|
||||
report(
|
||||
`Non-overload error on attempt ${modelAttempts}: ${modelError.message}`,
|
||||
'error'
|
||||
);
|
||||
throw modelError; // Re-throw non-overload errors immediately.
|
||||
}
|
||||
} // End inner catch
|
||||
} // End while loop
|
||||
|
||||
// If loop finished without getting information
|
||||
if (!additionalInformation) {
|
||||
// Only show debug info for text output (CLI)
|
||||
if (outputFormat === 'text') {
|
||||
console.log(
|
||||
'>>> DEBUG: additionalInformation is falsy! Value:',
|
||||
additionalInformation
|
||||
);
|
||||
if (outputFormat === 'text' && loadingIndicator) {
|
||||
// Stop indicator immediately since generateText is blocking
|
||||
stopLoadingIndicator(loadingIndicator);
|
||||
loadingIndicator = null;
|
||||
}
|
||||
throw new Error(
|
||||
'Failed to generate additional information after all attempts.'
|
||||
);
|
||||
}
|
||||
|
||||
// Only show debug info for text output (CLI)
|
||||
if (outputFormat === 'text') {
|
||||
console.log(
|
||||
'>>> DEBUG: Got additionalInformation:',
|
||||
additionalInformation.substring(0, 50) + '...'
|
||||
);
|
||||
}
|
||||
// Assign the result directly (generateTextService returns the text string)
|
||||
additionalInformation = streamResult ? streamResult.trim() : '';
|
||||
|
||||
if (!additionalInformation) {
|
||||
throw new Error('AI returned empty response.'); // Changed error message slightly
|
||||
}
|
||||
report(
|
||||
// Corrected log message to reflect generateText
|
||||
`Successfully generated text using AI role: ${serviceRole}.`,
|
||||
'info'
|
||||
);
|
||||
} catch (aiError) {
|
||||
report(`AI service call failed: ${aiError.message}`, 'error');
|
||||
throw aiError;
|
||||
} // Removed the inner finally block as streamingInterval is gone
|
||||
|
||||
// Create timestamp
|
||||
const currentDate = new Date();
|
||||
const timestamp = currentDate.toISOString();
|
||||
|
||||
// Format the additional information with timestamp
|
||||
const formattedInformation = `\n\n<info added on ${timestamp}>\n${additionalInformation}\n</info added on ${timestamp}>`;
|
||||
const formattedInformation = `\n\n<info added on ${currentDate.toISOString()}>\n${additionalInformation}\n</info added on ${currentDate.toISOString()}>`;
|
||||
|
||||
// Only show debug info for text output (CLI)
|
||||
if (outputFormat === 'text') {
|
||||
@@ -556,9 +376,9 @@ Provide concrete examples, code snippets, or implementation details when relevan
|
||||
' 1. Run task-master list --with-subtasks to see all available subtask IDs'
|
||||
);
|
||||
console.log(
|
||||
' 2. Use a valid subtask ID with the --id parameter in format \"parentId.subtaskId\"'
|
||||
' 2. Use a valid subtask ID with the --id parameter in format "parentId.subtaskId"'
|
||||
);
|
||||
} else if (error.message?.includes('empty response from AI')) {
|
||||
} else if (error.message?.includes('empty stream response')) {
|
||||
console.log(
|
||||
chalk.yellow(
|
||||
'\nThe AI model returned an empty response. This might be due to the prompt or API issues. Try rephrasing or trying again later.'
|
||||
@@ -575,11 +395,6 @@ Provide concrete examples, code snippets, or implementation details when relevan
|
||||
}
|
||||
|
||||
return null;
|
||||
} finally {
|
||||
// Final cleanup check for the indicator, although it should be stopped by now
|
||||
if (outputFormat === 'text' && loadingIndicator) {
|
||||
stopLoadingIndicator(loadingIndicator);
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
|
||||
Reference in New Issue
Block a user