← ClaudeAtlas

managing-contextlisted

Discovers and loads relevant project context from markdown documentation before each task. Matches context documents based on keywords, file paths, and task types. Use at task start to access project plans, architecture, and implementation status.
Open330/agt · ★ 1 · Data & Documents · score 62
Install: claude install-skill Open330/agt
# Context Manager Auto-load relevant project documentation. ## Context Directory Structure ``` context/ ├── planning/ # Roadmaps, implementation plans ├── architecture/ # System design, decisions ├── guides/ # Getting started, tutorials ├── operations/ # Deployment, troubleshooting └── README.md # Context overview ``` ## Workflow ### Step 1: Check for Context ```bash ls -la context/ 2>/dev/null || echo "No context directory" ``` ### Step 2: Match Documents Based on task keywords: - "monitoring" → `context/monitoring/` - "deploy" → `context/operations/` - "API" → `context/architecture/` ### Step 3: Load Relevant Docs Read top 3-5 matching documents: ```bash # Example matches for "add monitoring" context/monitoring/architecture.md (score: 0.92) context/operations/alerting.md (score: 0.85) ``` ### Step 4: Summarize Brief user on loaded context: ``` 📄 Loaded context: - monitoring/architecture.md: Prometheus + Grafana stack - operations/alerting.md: Alert rules and escalation ``` ## Matching Algorithm | Factor | Weight | |--------|--------| | Keyword in filename | 40% | | Keyword in category | 30% | | Task type match | 20% | | Recency | 10% | ## After Task: Update Context If significant work done: ```bash # Update implementation status echo "- [x] Feature X completed" >> context/planning/status.md ``` ## Best Practices **DO:** - Check context at task start - Update status after completing work - Keep docs concise **DON'T:** - Load enti