Ctx Collector

ctx-collector

Purpose

Aggregate task checkboxes from multiple task file formats across project repositories and maintain comprehensive project context reports. This skill operates in "no-clone mode"—it only reads from existing local directories under projects/, making it safe and fast for continuous scanning.

Key Features:

• Multi-format support: .specify/ (Speckit) and specs/ (context-planning) directories
• Project context analysis: Automatic tech stack, structure, and dependency extraction
• Auto-update system: Git hooks keep project reports synchronized
• Priority detection: P1/​P2/​P3 and [P] (Speckit) formats

When to Use This Skill

Use this skill when:

• The user requests to scan tasks across projects (e.g., /​ctx.scan)
• Building or refreshing the project backlog before planning
• Collecting task status updates from multiple repositories
• Needing a unified view of all open and completed tasks

Do not use this skill when:

• Working with a single project's tasks directly
• The user wants to modify tasks (this skill only reads)
• Projects are not structured under context/​projects/

How It Works

Task Collection Process

1. Discover task files recursively in projects/ directory:

   • Speckit format: .specify/ directories with task files
   • Context-planning format: specs/ directories with tasks.md
   • Optional: checklists/​**/​*.md files (if configured)
   • Works with any project structure you prefer

2. Parse checkboxes from discovered files:

   • Open tasks: - [ ] Task description
   • Completed tasks: - [X] Task description or - [x] Task description

3. Extract metadata from task lines and headings:

   • Task ID: Pattern T\d+ (e.g., T101, T250)
   • Priority: P1|P2|P3 or [P] (Speckit format, treated as P1)
   • Scope: Last two heading levels as context
   • File location and line number for traceability

Project Context Analysis

The skill also includes scripts/​analyze_project.py for comprehensive project analysis:

1. Overview: Extract project name and description from README/​pyproject.toml
2. Tech Stack: Detect languages and frameworks (Python, JavaScript/​TypeScript, Rust)
3. Structure: Analyze directory organization and key files
4. Entry Points: Identify CLI commands and main modules
5. Specs: Detect Speckit or context-planning task specifications
6. Tests: Discover test frameworks and coverage tools
7. Documentation: Index markdown and RST documentation files

Results are saved to reports/​projects/ as both Markdown and JSON files.

Auto-Update System

Git hooks automatically update project contexts on changes:

• post-commit: Updates changed projects after local commits
• post-merge: Updates changed projects after git pull
• post-checkout: Updates when switching branches

Install hooks with: bash scripts/​install_hooks.sh

See docs/​AUTO_UPDATE.md for detailed documentation.

1. Generate unified backlog at state/​backlog.yaml with structure:

   generated_at: "ISO-8601 timestamp"
   items:
     - uid: "my-project#T101"
       project: "my-project"
       file: "projects/​my-project/​specs/​tasks.md"
       line: 42
       id: "T101"
       title: "Task description"
       priority: "P1"
       status: "open"
       scope:
         section: "Feature Name"
         subsection: "Phase 1"
   

Note: Project names are derived from the directory structure automatically.

Using the Bundled Scripts

The skill includes two main scripts:

1. Task Scanner (scripts/​scan_tasks.py):

# Scan all tasks across projects
python3 skills/​ctx-collector/​scripts/​scan_tasks.py --verbose

The scanner automatically:

• Scans both .specify/ and specs/ directories
• Supports P1/​P2/​P3 and [P] priority formats
• Handles missing YAML library (falls back to JSON)
• Generates unique IDs for tasks without explicit IDs
• Creates state/ directory if needed
• Overwrites state/​backlog.yaml with fresh scan

2. Project Analyzer (scripts/​analyze_project.py):

# Analyze a specific project
python3 skills/​ctx-collector/​scripts/​analyze_project.py \
  projects/​my-org/​my-project \
  --output reports/​projects/​my-project.md

# Or use the auto-update system
bash scripts/​update_project_contexts.sh manual

The analyzer automatically:

• Extracts tech stack and dependencies
• Maps directory structure
• Finds entry points and CLI commands
• Detects test frameworks
• Indexes documentation
• Outputs both Markdown and JSON reports

Task ID and UID Generation

• Explicit IDs: Tasks with T\d+ pattern use that as ID
• Implicit UIDs: Tasks without IDs get hash-based UID from project#path:line:title
• Project prefix: All UIDs include project# for cross-project uniqueness

Priority Detection

Priority is determined in order:

1. Speckit format: [P] in task line is treated as P1 (highest priority)
2. Standard format: P1|P2|P3 marker in task line (e.g., - [ ] T101 P1 Fix critical bug)
3. Priority marker in nearest parent heading
4. Default to P2 if not specified

Examples:

- [ ] [P] Critical bug fix        # Treated as P1 (Speckit)
- [ ] T101 P1 High priority task  # P1 (standard)
- [ ] T102 P2 Normal task         # P2 (standard)
- [ ] T103 Regular task           # P2 (default)

Workflow Examples

Task Scanning (/​ctx.scan)

1. Verify projects/ contains your project directories
2. Execute scripts/​scan_tasks.py or implement inline scanning
3. Report summary: "Scanned X files, found Y tasks (Z open, W done)"
4. Confirm state/​backlog.yaml has been updated

Project Analysis (/​ctx.analyze <project>)

1. Verify project exists in projects/ directory
2. Execute scripts/​analyze_project.py for the specific project
3. Generate reports/​projects/<name>.md and <name>.json
4. Update reports/​projects/​INDEX.md with new project entry

Auto-Update on Git Operations

When you commit, pull, or checkout:

1. Git hook detects changed projects
2. scripts/​update_project_contexts.sh runs automatically
3. Updated reports are generated for changed projects
4. Deleted projects have their reports removed
5. INDEX.md is regenerated

Constraints

• Read-only operation: Never modifies source task files
• Local files only: No network access or git operations
• No cloning: Only processes existing local directories
• Stable output format: Other skills depend on backlog.yaml structure

Related Skills

• ctx-planning: Consumes state/​backlog.yaml to generate daily/​weekly plans
• Use /​ctx.update to modify planning parameters in BaseContext.yaml

Troubleshooting

If scanning produces unexpected results, check:

• Task files use proper checkbox syntax: - [ ] or - [X]
• Task IDs follow T\d+ pattern (e.g., T1, T100, not task-1)
• Priority markers use P1, P2, or P3 format
• Files are UTF-8 encoded
• state/ directory is writable