Cowork Export

Cowork Export Skill

Packages your local session context (git state, Qdrant memory, decisions, code changes) into a clipboard-ready briefing document that Claude on claude.ai can consume as a remote sub-agent.

Quick Start

# Full export: git state + Qdrant memory + task brief
python3 skills/cowork-export/scripts/export_context.py \
  --project <project-name> \
  --task "What the remote agent should do" \
  --output .tmp/cowork_briefing.md

# Minimal: git state only (no Qdrant dependency)
python3 skills/cowork-export/scripts/export_context.py \
  --git-only \
  --task "Review these changes and suggest improvements" \
  --output .tmp/cowork_briefing.md

# Include specific files as inline context
python3 skills/cowork-export/scripts/export_context.py \
  --project agi-agent-kit \
  --task "Refactor this module for clarity" \
  --include-files execution/memory_manager.py execution/session_boot.py \
  --output .tmp/cowork_briefing.md

# Copy to clipboard directly (macOS)
python3 skills/cowork-export/scripts/export_context.py \
  --project agi-agent-kit \
  --task "Write tests for the new feature" \
  --clipboard

How It Works

┌─────────────────────────────────────────────┐
│  1. Gather Context                          │
│     ├─ git branch, diff, recent commits     │
│     ├─ Qdrant: recent decisions & learnings │
│     └─ Specified files (inline snippets)    │
│                                             │
│  2. Build Briefing Document                 │
│     ├─ Project summary                      │
│     ├─ Current state (what was done)        │
│     ├─ Task assignment (what to do next)    │
│     └─ Constraints & guidelines             │
│                                             │
│  3. Export                                  │
│     ├─ .tmp/cowork_briefing.md (file)       │
│     ├─ --clipboard (macOS pbcopy)           │
│     └─ stdout (pipe-friendly)               │
└─────────────────────────────────────────────┘

Output Sections

The exported briefing contains:

Agent Protocol

When the user asks to export context to Cowork:

1. Ask what task the remote agent should perform (if not specified)
2. Ask which files to include inline (if the task requires code context)
3. Run the export script with appropriate flags
4. Read the output file to verify it looks correct
5. Copy to clipboard or show the user where the file is

When the user says "schedule" or "delegate" — add time-awareness:

python3 skills/cowork-export/scripts/export_context.py \
  --project myapp \
  --task "Run the full test suite and report failures" \
  --deadline "2026-03-23" \
  --priority high \
  --output .tmp/cowork_briefing.md

Project Bootstrap Pattern

When the user wants Cowork to create a full new project (not just review existing code):

1. Ask for project specs — Name, purpose, tech stack, automation goals
2. Export a bootstrap briefing with framework structure embedded:

python3 skills/cowork-export/scripts/export_context.py \
  --project new-project-name \
  --task "Create a complete Python automation project with these specifications: [specs]. Use the AGI Agent Kit structure: directives/ for SOPs, execution/ for scripts, skills/ for modular capabilities. Include a CLAUDE.md with agent instructions, a session_boot.py, and a memory_manager.py integration." \
  --include-files CLAUDE.md execution/session_boot.py \
  --constraints "Follow the 3-layer architecture: directives (intent), orchestration (agent), execution (scripts). All scripts must accept CLI args and return JSON." \
  --priority high \
  --clipboard

3. Paste into Cowork (claude.ai or Desktop) → Cowork builds the full project
4. Pull results back → Copy from Cowork's folder into your repo
5. Record the handoff so local agents know

Full Project Ideas via Cowork

Common automation projects to delegate:

• Email automation pipeline (Gmail MCP → classify → respond/archive)
• Competitor monitoring dashboard (scrape → compare → HTML report)
• Social media content scheduler (draft → review → post via API)
• Data pipeline (ingest → transform → export to Google Sheets)
• Internal tool (Slack bot → process requests → update database)

Cross-Agent Integration

After exporting, store a handoff record so local agents know work was delegated:

python3 execution/cross_agent_context.py handoff \
  --from "claude" --to "cowork" \
  --task "<task description>" \
  --project <project-name>

Flags Reference