ros/claude-task-master
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

feat: Centralize AI prompts into JSON templates (#882)

Browse Source 
* centralize prompt management

* add changeset

* add variant key to determine prompt version

* update tests and add prompt manager test

* determine internal path, don't use projectRoot

* add promptManager mock

* detailed prompt docs

* add schemas and validator packages

* add validate prompts command

* add schema validation

* update tests

* move schemas to src/prompts/schemas

* use this.promptsDir for better semantics

* add prompt schemas

* version schema files & update links

* remove validate command

* expect dependencies

* update docs

* fix test

* remove suggestmode to ensure clean keys

* remove default variant from research and update schema

* now handled by prompt manager

* add manual test to verify prompts

* remove incorrect batch variant

* consolidate variants

* consolidate analyze-complexity to just default variant

* consolidate parse-prd variants

* add eq handler for handlebars

* consolidate research prompt variants

* use brevity

* consolidate variants for update subtask

* add not handler

* consolidate variants for update-task

* consolidate update-tasks variants

* add conditional content to prompt when research used

* update prompt tests

* show correct research variant

* make variant names link to below

* remove changset

* restore gitignore

* Merge branch 'next' of https://github.com/eyaltoledano/claude-task-master into joedanz/centralize-prompts

# Conflicts:
#	package-lock.json
#	scripts/modules/task-manager/expand-task.js
#	scripts/modules/task-manager/parse-prd.js

remove unused

* add else

* update tests

* update biome optional dependencies

* responsive html output for mobile
This commit is contained in:
Joe Danziger
2025-07-10 03:52:11 -04:00
committed by GitHub
parent 4bc8029080
commit a65ad0a47c
36 changed files with 6180 additions and 9034 deletions
Download Patch File Download Diff File Expand all files Collapse all files

93
scripts/modules/task-manager/parse-prd.js
View File

@@ -18,6 +18,7 @@ import {
import { generateObjectService } from '../ai-services-unified.js';
import { getDebugFlag } from '../config-manager.js';
import { getPromptManager } from '../prompt-manager.js';
import generateTaskFiles from './generate-task-files.js';
import { displayAiUsageSummary } from '../ui.js';
@@ -147,10 +148,8 @@ async function parsePRD(prdPath, tasksPath, numTasks, options = {}) {
report(overwriteError.message, 'error');
if (outputFormat === 'text') {
console.error(chalk.red(overwriteError.message));
process.exit(1);
} else {
throw overwriteError;
}
throw overwriteError;
} else {
// Force overwrite is true
report(
@@ -172,74 +171,24 @@ async function parsePRD(prdPath, tasksPath, numTasks, options = {}) {
throw new Error(`Input file ${prdPath} is empty or could not be read.`);
}
// Research-specific enhancements to the system prompt
const researchPromptAddition = research
? `\nBefore breaking down the PRD into tasks, you will:
1. Research and analyze the latest technologies, libraries, frameworks, and best practices that would be appropriate for this project
2. Identify any potential technical challenges, security concerns, or scalability issues not explicitly mentioned in the PRD without discarding any explicit requirements or going overboard with complexity -- always aim to provide the most direct path to implementation, avoiding over-engineering or roundabout approaches
3. Consider current industry standards and evolving trends relevant to this project (this step aims to solve LLM hallucinations and out of date information due to training data cutoff dates)
4. Evaluate alternative implementation approaches and recommend the most efficient path
5. Include specific library versions, helpful APIs, and concrete implementation guidance based on your research
6. Always aim to provide the most direct path to implementation, avoiding over-engineering or roundabout approaches
// Load prompts using PromptManager
const promptManager = getPromptManager();
Your task breakdown should incorporate this research, resulting in more detailed implementation guidance, more accurate dependency mapping, and more precise technology recommendations than would be possible from the PRD text alone, while maintaining all explicit requirements and best practices and all details and nuances of the PRD.`
: '';
// Get defaultTaskPriority from config
const { getDefaultPriority } = await import('../config-manager.js');
const defaultTaskPriority = getDefaultPriority(projectRoot) || 'medium';
// Base system prompt for PRD parsing
const systemPrompt = `You are an AI assistant specialized in analyzing Product Requirements Documents (PRDs) and generating a structured, logically ordered, dependency-aware and sequenced list of development tasks in JSON format.${researchPromptAddition}
Analyze the provided PRD content and generate ${numTasks > 0 ? 'approximately ' + numTasks : 'an appropriate number of'} top-level development tasks. If the complexity or the level of detail of the PRD is high, generate more tasks relative to the complexity of the PRD
Each task should represent a logical unit of work needed to implement the requirements and focus on the most direct and effective way to implement the requirements without unnecessary complexity or overengineering. Include pseudo-code, implementation details, and test strategy for each task. Find the most up to date information to implement each task.
Assign sequential IDs starting from ${nextId}. Infer title, description, details, and test strategy for each task based *only* on the PRD content.
Set status to 'pending', dependencies to an empty array [], and priority to 'medium' initially for all tasks.
Respond ONLY with a valid JSON object containing a single key "tasks", where the value is an array of task objects adhering to the provided Zod schema. Do not include any explanation or markdown formatting.
Each task should follow this JSON structure:
{
"id": number,
"title": string,
"description": string,
"status": "pending",
"dependencies": number[] (IDs of tasks this depends on),
"priority": "high" | "medium" | "low",
"details": string (implementation details),
"testStrategy": string (validation approach)
}
Guidelines:
1. ${numTasks > 0 ? 'Unless complexity warrants otherwise' : 'Depending on the complexity'}, create ${numTasks > 0 ? 'exactly ' + numTasks : 'an appropriate number of'} tasks, numbered sequentially starting from ${nextId}
2. Each task should be atomic and focused on a single responsibility following the most up to date best practices and standards
3. Order tasks logically - consider dependencies and implementation sequence
4. Early tasks should focus on setup, core functionality first, then advanced features
5. Include clear validation/testing approach for each task
6. Set appropriate dependency IDs (a task can only depend on tasks with lower IDs, potentially including existing tasks with IDs less than ${nextId} if applicable)
7. Assign priority (high/medium/low) based on criticality and dependency order
8. Include detailed implementation guidance in the "details" field${research ? ', with specific libraries and version recommendations based on your research' : ''}
9. If the PRD contains specific requirements for libraries, database schemas, frameworks, tech stacks, or any other implementation details, STRICTLY ADHERE to these requirements in your task breakdown and do not discard them under any circumstance
10. Focus on filling in any gaps left by the PRD or areas that aren't fully specified, while preserving all explicit requirements
11. Always aim to provide the most direct path to implementation, avoiding over-engineering or roundabout approaches${research ? '\n12. For each task, include specific, actionable guidance based on current industry standards and best practices discovered through research' : ''}`;
// Build user prompt with PRD content
const userPrompt = `Here's the Product Requirements Document (PRD) to break down into approximately ${numTasks > 0 ? 'approximately ' + numTasks : 'an appropriate number of'} tasks, starting IDs from ${nextId}:${research ? '\n\nRemember to thoroughly research current best practices and technologies before task breakdown to provide specific, actionable implementation details.' : ''}\n\n${prdContent}\n\n
Return your response in this format:
{
"tasks": [
{
"id": 1,
"title": "Setup Project Repository",
"description": "...",
...
},
...
],
"metadata": {
"projectName": "PRD Implementation",
"totalTasks": {number of tasks},
"sourceFile": "${prdPath}",
"generatedAt": "YYYY-MM-DD"
}
}`;
const { systemPrompt, userPrompt } = await promptManager.loadPrompt(
'parse-prd',
{
research,
numTasks,
nextId,
prdContent,
prdPath,
defaultTaskPriority
}
);
// Call the unified AI service
report(
@@ -420,11 +369,9 @@ Guidelines:
// Use projectRoot for debug flag check
console.error(error);
}
process.exit(1);
} else {
throw error; // Re-throw for JSON output
}
throw error; // Always re-throw for proper error handling
}
}