mirror of
https://github.com/catlog22/Claude-Code-Workflow.git
synced 2026-02-14 02:42:04 +08:00
refactor: convert context-gather to agent-driven execution and fix path mismatch
- Refactor context-gather.md to use general-purpose agent delegation - Change from 4-phase manual execution to 2-phase agent-driven flow - Move project structure analysis and documentation loading to agent execution - Add Step 0: Foundation Setup for agent to execute first - Update agent context passing to minimal configuration - Add MCP tools integration guidance for agent - Fix critical path mismatch in workflow data flow - Update plan.md Phase 2 output path from .context/ to .process/ - Align with context-gather.md output location (.process/context-package.json) - Ensure correct data flow: context-gather → concept-enhanced - Update concept-enhanced.md line selection (minor formatting) Verified path consistency across all workflow commands: - context-gather.md outputs to .process/ - concept-enhanced.md reads from .process/ - plan.md passes correct .process/ path - All workflow tools now use consistent .process/ directory 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com>
This commit is contained in:
catlog22
@@ -1,6 +1,6 @@
|
||||
---
|
||||
name: gather
|
||||
description: Intelligently collect project context based on task description and package into standardized JSON
|
||||
description: Intelligently collect project context using general-purpose agent based on task description and package into standardized JSON
|
||||
argument-hint: "--session WFS-session-id \"task description\""
|
||||
examples:
|
||||
- /workflow:tools:context-gather --session WFS-user-auth "Implement user authentication system"
|
||||
@@ -11,25 +11,105 @@ examples:
|
||||
# Context Gather Command (/workflow:tools:context-gather)
|
||||
|
||||
## Overview
|
||||
Intelligent context collector that gathers relevant information from project codebase, documentation, and dependencies based on task descriptions, generating standardized context packages.
|
||||
Agent-driven intelligent context collector that gathers relevant information from project codebase, documentation, and dependencies based on task descriptions, generating standardized context packages.
|
||||
|
||||
## Core Philosophy
|
||||
- **Agent-Driven**: Delegate execution to general-purpose agent for autonomous operation
|
||||
- **Two-Phase Flow**: Discovery (context loading) → Execution (context gathering and packaging)
|
||||
- **Memory-First**: Reuse loaded documents from conversation memory
|
||||
- **MCP-Enhanced**: Use MCP tools for advanced code analysis and file discovery
|
||||
- **Intelligent Collection**: Auto-identify relevant resources based on keyword analysis
|
||||
- **Comprehensive Coverage**: Collect code, documentation, configurations, and dependencies
|
||||
- **Standardized Output**: Generate unified format context-package.json
|
||||
- **Efficient Execution**: Optimize collection strategies to avoid irrelevant information
|
||||
|
||||
## Core Responsibilities
|
||||
- **Keyword Extraction**: Extract core keywords from task descriptions
|
||||
- **Smart Documentation Loading**: Load relevant project documentation based on keywords
|
||||
- **Code Structure Analysis**: Analyze project structure to locate relevant code files
|
||||
- **Dependency Discovery**: Identify tech stack and dependency relationships
|
||||
- **MCP Tools Integration**: Leverage code-index tools for enhanced collection
|
||||
- **Context Packaging**: Generate standardized JSON context packages
|
||||
## Execution Lifecycle
|
||||
|
||||
## Execution Process
|
||||
### Phase 1: Discovery & Context Loading
|
||||
**⚡ Memory-First Rule**: Skip file loading if documents already in conversation memory
|
||||
|
||||
### Phase 1: Task Analysis
|
||||
**Agent Context Package**:
|
||||
```javascript
|
||||
{
|
||||
"session_id": "WFS-[session-id]",
|
||||
"task_description": "[user provided task description]",
|
||||
"session_metadata": {
|
||||
// If in memory: use cached content
|
||||
// Else: Load from .workflow/{session-id}/workflow-session.json
|
||||
},
|
||||
"mcp_capabilities": {
|
||||
// Agent will use these tools to discover project context
|
||||
"code_index": true,
|
||||
"exa_code": true,
|
||||
"exa_web": true
|
||||
}
|
||||
}
|
||||
|
||||
// Agent will autonomously execute:
|
||||
// - Project structure analysis: bash(~/.claude/scripts/get_modules_by_depth.sh)
|
||||
// - Documentation loading: Read(CLAUDE.md), Read(README.md)
|
||||
```
|
||||
|
||||
**Discovery Actions**:
|
||||
1. **Load Session Context** (if not in memory)
|
||||
```javascript
|
||||
if (!memory.has("workflow-session.json")) {
|
||||
Read(.workflow/{session-id}/workflow-session.json)
|
||||
}
|
||||
```
|
||||
|
||||
### Phase 2: Agent Execution (Context Gathering & Packaging)
|
||||
|
||||
**Agent Invocation**:
|
||||
```javascript
|
||||
Task(
|
||||
subagent_type="general-purpose",
|
||||
description="Gather project context and generate context package",
|
||||
prompt=`
|
||||
## Execution Context
|
||||
|
||||
**Session ID**: WFS-{session-id}
|
||||
**Task Description**: {task_description}
|
||||
**Mode**: Agent-Driven Context Gathering
|
||||
|
||||
## Phase 1: Discovery Results (Provided Context)
|
||||
|
||||
### Session Metadata
|
||||
{session_metadata_content}
|
||||
|
||||
### MCP Capabilities
|
||||
- code-index: Available for file discovery and code search
|
||||
- exa-code: Available for external research
|
||||
- exa-web: Available for web search
|
||||
|
||||
## Phase 2: Context Gathering Task
|
||||
|
||||
### Core Responsibilities
|
||||
1. **Project Structure Analysis**: Execute get_modules_by_depth.sh for architecture overview
|
||||
2. **Documentation Loading**: Load CLAUDE.md, README.md and relevant documentation
|
||||
3. **Keyword Extraction**: Extract core keywords from task description
|
||||
4. **Smart File Discovery**: Use MCP code-index tools to locate relevant files
|
||||
5. **Code Structure Analysis**: Analyze project structure to identify relevant modules
|
||||
6. **Dependency Discovery**: Identify tech stack and dependency relationships
|
||||
7. **Context Packaging**: Generate standardized JSON context package
|
||||
|
||||
### Execution Process
|
||||
|
||||
#### Step 0: Foundation Setup (Execute First)
|
||||
1. **Project Structure Analysis**
|
||||
Execute to get comprehensive architecture overview:
|
||||
\`\`\`javascript
|
||||
bash(~/.claude/scripts/get_modules_by_depth.sh)
|
||||
\`\`\`
|
||||
|
||||
2. **Load Project Documentation** (if not in memory)
|
||||
Load core project documentation:
|
||||
\`\`\`javascript
|
||||
Read(CLAUDE.md)
|
||||
Read(README.md)
|
||||
// Load other relevant documentation based on session context
|
||||
\`\`\`
|
||||
|
||||
#### Step 1: Task Analysis
|
||||
1. **Keyword Extraction**
|
||||
- Parse task description to extract core keywords
|
||||
- Identify technical domain (auth, API, frontend, backend, etc.)
|
||||
@@ -40,23 +120,31 @@ Intelligent context collector that gathers relevant information from project cod
|
||||
- Identify potentially involved modules and components
|
||||
- Set file type filters
|
||||
|
||||
### Phase 2: Project Structure Exploration
|
||||
1. **Architecture Analysis**
|
||||
- Use `~/.claude/scripts/get_modules_by_depth.sh` for comprehensive project structure
|
||||
- Analyze project layout and module organization
|
||||
- Identify key directories and components
|
||||
#### Step 2: MCP-Enhanced File Discovery
|
||||
1. **Code File Location**
|
||||
Use MCP code-index tools:
|
||||
\`\`\`javascript
|
||||
// Find files by pattern
|
||||
mcp__code-index__find_files(pattern="*{keyword}*")
|
||||
|
||||
2. **Code File Location**
|
||||
- Use MCP tools for precise search: `mcp__code-index__find_files()` and `mcp__code-index__search_code_advanced()`
|
||||
- Search for relevant source code files based on keywords
|
||||
- Locate implementation files, interfaces, and modules
|
||||
// Search code content
|
||||
mcp__code-index__search_code_advanced(
|
||||
pattern="{keyword_patterns}",
|
||||
file_pattern="*.{ts,js,py,go,md}",
|
||||
context_lines=3
|
||||
)
|
||||
|
||||
3. **Documentation Collection**
|
||||
- Load CLAUDE.md and README.md
|
||||
- Load relevant documentation from .workflow/docs/ based on keywords
|
||||
- Collect configuration files (package.json, requirements.txt, etc.)
|
||||
// Get file summaries
|
||||
mcp__code-index__get_file_summary(file_path="relevant/file.ts")
|
||||
\`\`\`
|
||||
|
||||
### Phase 3: Intelligent Filtering & Association
|
||||
2. **Configuration Files Discovery**
|
||||
Locate: package.json, requirements.txt, Cargo.toml, tsconfig.json, etc.
|
||||
|
||||
3. **Test Files Location**
|
||||
Find test files related to task keywords
|
||||
|
||||
#### Step 3: Intelligent Filtering & Association
|
||||
1. **Relevance Scoring**
|
||||
- Score based on keyword match degree
|
||||
- Score based on file path relevance
|
||||
@@ -67,17 +155,15 @@ Intelligent context collector that gathers relevant information from project cod
|
||||
- Identify inter-module dependencies
|
||||
- Determine core and optional dependencies
|
||||
|
||||
### Phase 4: Context Packaging
|
||||
1. **Standardized Output**
|
||||
- Generate context-package.json
|
||||
- Organize resources by type and importance
|
||||
- Add relevance descriptions and usage recommendations
|
||||
#### Step 4: Context Packaging
|
||||
Generate standardized context-package.json following the format below
|
||||
|
||||
## Context Package Format
|
||||
### Required Output
|
||||
|
||||
Generated context package format:
|
||||
**Output Location**: \`.workflow/{session-id}/.process/context-package.json\`
|
||||
|
||||
```json
|
||||
**Output Format**:
|
||||
\`\`\`json
|
||||
{
|
||||
"metadata": {
|
||||
"task_description": "Implement user authentication system",
|
||||
@@ -138,163 +224,130 @@ Generated context package format:
|
||||
"test_files": 1
|
||||
}
|
||||
}
|
||||
```
|
||||
\`\`\`
|
||||
|
||||
## MCP Tools Integration
|
||||
### Quality Validation
|
||||
|
||||
### Code Index Integration
|
||||
```bash
|
||||
# Set project path
|
||||
Before completion, verify:
|
||||
- [ ] context-package.json created in correct location
|
||||
- [ ] Valid JSON format with all required fields
|
||||
- [ ] Metadata includes task description, keywords, complexity
|
||||
- [ ] Assets array contains relevant files with priorities
|
||||
- [ ] Tech stack accurately identified
|
||||
- [ ] Statistics section provides file counts
|
||||
- [ ] File relevance accuracy >80%
|
||||
- [ ] No sensitive information exposed
|
||||
|
||||
### Performance Optimization
|
||||
|
||||
**Large Project Optimization**:
|
||||
- File count limit: Maximum 50 files per type
|
||||
- Size filtering: Skip oversized files (>10MB)
|
||||
- Depth limit: Maximum search depth of 3 levels
|
||||
- Use MCP tools for efficient discovery
|
||||
|
||||
**MCP Tools Integration**:
|
||||
Agent should use MCP code-index tools when available:
|
||||
\`\`\`javascript
|
||||
// Set project path
|
||||
mcp__code-index__set_project_path(path="{current_project_path}")
|
||||
|
||||
# Refresh index to ensure latest
|
||||
// Refresh index
|
||||
mcp__code-index__refresh_index()
|
||||
|
||||
# Search relevant files
|
||||
// Find files by pattern
|
||||
mcp__code-index__find_files(pattern="*{keyword}*")
|
||||
|
||||
# Search code content
|
||||
// Search code content
|
||||
mcp__code-index__search_code_advanced(
|
||||
pattern="{keyword_patterns}",
|
||||
file_pattern="*.{ts,js,py,go,md}",
|
||||
context_lines=3
|
||||
)
|
||||
\`\`\`
|
||||
|
||||
**Fallback Strategy**:
|
||||
When MCP tools unavailable, agent should use traditional commands:
|
||||
- \`find\` for file discovery
|
||||
- \`rg\` or \`grep\` for content search
|
||||
- Bash commands from project structure analysis
|
||||
|
||||
## Output
|
||||
|
||||
Generate context-package.json and report completion:
|
||||
- Task description: {description}
|
||||
- Keywords extracted: {count}
|
||||
- Files collected: {total}
|
||||
- Source files: {count}
|
||||
- Documentation: {count}
|
||||
- Configuration: {count}
|
||||
- Tests: {count}
|
||||
- Tech stack identified: {frameworks/libraries}
|
||||
- Output location: .workflow/{session-id}/.process/context-package.json
|
||||
\`
|
||||
)
|
||||
\`\`\`
|
||||
|
||||
## Command Integration
|
||||
|
||||
### Usage
|
||||
```bash
|
||||
# Basic usage
|
||||
/workflow:tools:context-gather --session WFS-auth "Implement JWT authentication"
|
||||
|
||||
# Called by /workflow:plan
|
||||
SlashCommand(command="/workflow:tools:context-gather --session WFS-[id] \\"[task description]\\"")
|
||||
```
|
||||
|
||||
### Agent Context Passing
|
||||
|
||||
**Memory-Aware Context Assembly**:
|
||||
```javascript
|
||||
// Assemble minimal context package for agent
|
||||
// Agent will execute project structure analysis and documentation loading
|
||||
const agentContext = {
|
||||
session_id: "WFS-[id]",
|
||||
task_description: "[user provided task description]",
|
||||
|
||||
// Use memory if available, else load
|
||||
session_metadata: memory.has("workflow-session.json")
|
||||
? memory.get("workflow-session.json")
|
||||
: Read(.workflow/WFS-[id]/workflow-session.json),
|
||||
|
||||
// MCP capabilities - agent will use these tools
|
||||
mcp_capabilities: {
|
||||
code_index: true,
|
||||
exa_code: true,
|
||||
exa_web: true
|
||||
}
|
||||
}
|
||||
|
||||
// Note: Agent will execute these steps autonomously:
|
||||
// - bash(~/.claude/scripts/get_modules_by_depth.sh) for project structure
|
||||
// - Read(CLAUDE.md) and Read(README.md) for documentation
|
||||
```
|
||||
|
||||
## Session ID Integration
|
||||
|
||||
### Session ID Usage
|
||||
- **Required Parameter**: `--session WFS-session-id`
|
||||
- **Session Context Loading**: Load existing session state and task summaries
|
||||
- **Session Continuity**: Maintain context across pipeline phases
|
||||
- **Session Context Loading**: Load existing session state and metadata
|
||||
- **Session Continuity**: Maintain context across workflow pipeline phases
|
||||
|
||||
### Session State Management
|
||||
```bash
|
||||
# Validate session exists
|
||||
if [ ! -d ".workflow/${session_id}" ]; then
|
||||
echo "❌ Session ${session_id} not found"
|
||||
exit 1
|
||||
fi
|
||||
|
||||
# Load session metadata
|
||||
session_metadata=".workflow/${session_id}/workflow-session.json"
|
||||
```
|
||||
|
||||
## Output Location
|
||||
|
||||
Context package output location:
|
||||
```
|
||||
.workflow/{session_id}/.process/context-package.json
|
||||
```
|
||||
|
||||
## Error Handling
|
||||
|
||||
### Common Error Handling
|
||||
1. **No Active Session**: Create temporary session directory
|
||||
2. **MCP Tools Unavailable**: Fallback to traditional bash commands
|
||||
3. **Permission Errors**: Prompt user to check file permissions
|
||||
4. **Large Project Optimization**: Limit file count, prioritize high-relevance files
|
||||
|
||||
### Graceful Degradation Strategy
|
||||
```bash
|
||||
# Fallback when MCP unavailable
|
||||
if ! command -v mcp__code-index__find_files; then
|
||||
# Use find command for file discovery
|
||||
find . -name "*{keyword}*" -type f -not -path "*/node_modules/*" -not -path "*/.git/*"
|
||||
|
||||
# Alternative pattern matching
|
||||
find . -type f \( -name "*.ts" -o -name "*.js" -o -name "*.py" -o -name "*.go" \) -exec grep -l "{keyword}" {} \;
|
||||
fi
|
||||
|
||||
# Use ripgrep instead of MCP search
|
||||
rg "{keywords}" --type-add 'source:*.{ts,js,py,go}' -t source --max-count 30
|
||||
|
||||
# Content-based search with context
|
||||
rg -A 3 -B 3 "{keywords}" --type-add 'source:*.{ts,js,py,go}' -t source
|
||||
|
||||
# Quick relevance check
|
||||
grep -r --include="*.{ts,js,py,go}" -l "{keywords}" . | head -15
|
||||
|
||||
# Test files discovery
|
||||
find . -name "*test*" -o -name "*spec*" | grep -E "\.(ts|js|py|go)$" | head -10
|
||||
|
||||
# Import/dependency analysis
|
||||
rg "^(import|from|require|#include)" --type-add 'source:*.{ts,js,py,go}' -t source | head -20
|
||||
```
|
||||
|
||||
## Performance Optimization
|
||||
|
||||
### Large Project Optimization Strategy
|
||||
- **File Count Limit**: Maximum 50 files per type
|
||||
- **Size Filtering**: Skip oversized files (>10MB)
|
||||
- **Depth Limit**: Maximum search depth of 3 levels
|
||||
- **Caching Strategy**: Cache project structure analysis results
|
||||
|
||||
### Parallel Processing
|
||||
- Documentation collection and code search in parallel
|
||||
- MCP tool calls and traditional commands in parallel
|
||||
- Reduce I/O wait time
|
||||
|
||||
## Essential Bash Commands (Max 10)
|
||||
|
||||
### 1. Project Structure Analysis
|
||||
```bash
|
||||
~/.claude/scripts/get_modules_by_depth.sh
|
||||
```
|
||||
|
||||
### 2. File Discovery by Keywords
|
||||
```bash
|
||||
find . -name "*{keyword}*" -type f -not -path "*/node_modules/*" -not -path "*/.git/*"
|
||||
```
|
||||
|
||||
### 3. Content Search in Code Files
|
||||
```bash
|
||||
rg "{keyword}" --type-add 'source:*.{ts,js,py,go}' -t source --max-count 20
|
||||
```
|
||||
|
||||
### 4. Configuration Files Discovery
|
||||
```bash
|
||||
find . -maxdepth 3 \( -name "*.json" -o -name "package.json" -o -name "requirements.txt" -o -name "Cargo.toml" \) -not -path "*/node_modules/*"
|
||||
```
|
||||
|
||||
### 5. Documentation Files Collection
|
||||
```bash
|
||||
find . -name "*.md" -o -name "README*" -o -name "CLAUDE.md" | grep -v node_modules | head -10
|
||||
```
|
||||
|
||||
### 6. Test Files Location
|
||||
```bash
|
||||
find . \( -name "*test*" -o -name "*spec*" \) -type f | grep -E "\.(js|ts|py|go)$" | head -10
|
||||
```
|
||||
|
||||
### 7. Function/Class Definitions Search
|
||||
```bash
|
||||
rg "^(function|def|func|class|interface)" --type-add 'source:*.{ts,js,py,go}' -t source -n --max-count 15
|
||||
```
|
||||
|
||||
### 8. Import/Dependency Analysis
|
||||
```bash
|
||||
rg "^(import|from|require|#include)" --type-add 'source:*.{ts,js,py,go}' -t source | head -15
|
||||
```
|
||||
|
||||
### 9. Workflow Session Information
|
||||
```bash
|
||||
find .workflow/ -name "*.json" -path "*/${session_id}/*" -o -name "workflow-session.json" | head -5
|
||||
```
|
||||
|
||||
### 10. Context-Aware Content Search
|
||||
```bash
|
||||
rg -A 2 -B 2 "{keywords}" --type-add 'source:*.{ts,js,py,go}' -t source --max-count 10
|
||||
### Session Validation
|
||||
```javascript
|
||||
// Validate session exists
|
||||
const sessionPath = `.workflow/${session_id}`;
|
||||
if (!fs.existsSync(sessionPath)) {
|
||||
console.error(`❌ Session ${session_id} not found`);
|
||||
process.exit(1);
|
||||
}
|
||||
```
|
||||
|
||||
## Success Criteria
|
||||
- Generate valid context-package.json file
|
||||
- Contains sufficient relevant information for subsequent analysis
|
||||
- Execution time controlled within 30 seconds
|
||||
- File relevance accuracy rate >80%
|
||||
- Valid context-package.json generated in correct location
|
||||
- Contains sufficient relevant information (>80% relevance)
|
||||
- Execution completes within reasonable time (<2 minutes)
|
||||
- All required fields present and properly formatted
|
||||
- Agent reports completion status with statistics
|
||||
|
||||
## Related Commands
|
||||
- `/workflow:tools:concept-enhanced` - Consumes output of this command for analysis
|
||||
- `/workflow:plan` - Calls this command to gather context
|
||||
- `/workflow:status` - Can display context collection status
|
||||
Reference in New Issue
Block a user