agentic-engineering-framework: Command for Claude Code

/capture - Emergency Conversation Capture

When the user types /capture, execute this workflow to save the current conversation to disk before it is lost.

Step 1: Check active task

Read .context/working/focus.yaml to get current_task.

If no active task or current_task is empty:

• Print: "No active task. Run fw work-on 'topic' first to create one, then retry /capture."
• Stop here.

Step 2: Extract conversation from transcript

Run:

python3 agents/capture/read-transcript.py

This reads the live JSONL transcript and returns JSON with:

• turns: array of {role, content, timestamp} objects
• captured_turns: how many turns were captured
• capture_mode: how the boundary was detected

If the script exits non-zero or returns empty turns, print the warning from stderr and stop.

Step 3: Write the artifact

Generate the artifact path:

docs/reports/{current_task}-capture-{YYYY-MM-DD-HH}.md

Write the file with this structure:

# {current_task}: Conversation Capture — {YYYY-MM-DD HH:MM}

> Captured via /capture | Turns: {captured_turns} | Mode: {capture_mode}
> Task: {current_task}

## Topic / Problem Statement

{Summarise in 2-3 sentences what this conversation is about, based on the turns.}

## Key Insights

{List the most important findings or conclusions reached, as bullet points.
Extract from the conversation content — don't hallucinate.}

## Options Explored

{If alternatives were discussed, list them with brief notes on each.
Skip this section if no options were compared.}

## Decisions Made

{List any explicit decisions or directions agreed on.
Skip if no decisions were made.}

## Open Questions

{List anything that was raised but not resolved.}

## Conversation Log

{For each turn in the turns array, format as:}

**Human** _{timestamp}_
{content}

---

**Agent** _{timestamp}_
{content}

---

Step 4: Commit

Run:

fw git commit -m "{current_task}: /capture — conversation artifact"

Step 5: Report back