load-context

Load Session

Menu-driven session resume for non-CLI interfaces (Cowork desktop, Telegram, CC CLI). Reads CONTEXT files written by both /save-context (cowork) and /save-context (dstoic) — same format.

Phase 1: Scan Available Sessions

Run both Glob calls in parallel (single tool message):

Glob: CONTEXT-*-llm.md (current dir)
Glob: done/CONTEXT-*-llm.md (archived)

For each file found, Read first 10 lines to extract header:

• stream name (from filename: CONTEXT-{name}-llm.md → {name}, CONTEXT-llm.md → default)
• status
• focus
• saved date
• goal (if present)

⚠️ AskUserQuestion Guard

After EVERY AskUserQuestion call, check if answers are empty/blank. If empty: output "⚠️ Questions didn't display (known bug).", present options as numbered text list, WAIT for user reply.

Phase 2: Present Menu

Sort by saved date (most recent first). Separate active from archived.

📂 Available sessions:

📍 Active
  1. 🔄 {stream-1} ({relative-date}, {status})
     → {focus}
  2. 🏗️ {stream-2} ({relative-date}, {status})
     → {focus}

📦 Archived
  3. ✅ {stream-3} ({relative-date}, {status})
     → {focus}

Resume which? (number) or 0 to start fresh

If only 1 session exists → still show menu but suggest it: Resume **{stream}**? (yes / no)

If 0 sessions found → No saved sessions found. Use /switch to select a project, or start working and /save when ready.

Wait for user reply.

Phase 3: Load Selected Session

If user picks a number → Read the full CONTEXT file. If user picks 0 → respond: Starting fresh. What would you like to work on?

If session was from done/ → note: 📦 This session was archived (status: {status}). Resuming will create a new active session.

Parse CONTEXT file sections:

• Header (stream, saved, status, focus, goal)
• Project
• Next Tasks
• Session (progression, decisions, thinking, unexpected)
• Hot Files
• Refs

Phase 4: Present Resume Report

📍 Resumed: {stream name}
⏰ Last saved: {date, relative}
🎯 Focus: {focus}
🎯 Goal: {goal}

📋 Where you left off:
  {session progression — 2-3 key bullets}

🔑 Decisions made:
  {decisions — bullets if any}

✅ Next tasks:
  1. {task 1}
  2. {task 2}
  3. {task 3}

🔥 Key files:
  - {file}: {role}

What would you like to tackle first?

Graceful degradation: missing sections → skip silently. Only Header + Next Tasks required. Token budget: report under 1500 tokens. Summarize, don't dump.

Phase 5: Offer Expansion (optional)

If the session has OpenSpec references or >5 hot files:

💡 Want me to load the full context? (reads top hot files + OpenSpec specs)
Reply yes for deep context, or just start working.

Rules

1. Menu-first — always show available sessions as numbered choices
2. Human-friendly dates — "2 weeks ago", "yesterday", not ISO timestamps
3. Focus line as preview — always show it under menu entry
4. Graceful on empty — no sessions = helpful guidance, not error
5. Client-agnostic output — works in terminal, Cowork, Telegram. Simple lists + emoji
6. No shell scripts — pure tool calls only (Glob, Read, AskUserQuestion)
7. Token budget — resume report under 1500 tokens