A2a Send

Parse $ARGUMENTS into:

• URL_OR_ALIAS: first non-flag token
• MESSAGE: all remaining non-flag tokens joined as the message string
• BACKGROUND: true if --background present, false otherwise
• STREAM: true if --stream present
• TIMEOUT: value after --timeout if present (e.g. 60s, 2m)
• TASK_ID: value after --task if present
• CONTEXT_ID: value after --context if present

Process

1. Check prerequisites: Verify that a2a is installed before proceeding:

   if ! command -v a2a &>/dev/null; then
     echo "Error: 'a2a' CLI not found. Install with:"
     echo "  go install github.com/a2aproject/a2a-go/v2/cmd/a2a@latest"
     exit 1
   fi
   

2. Resolve alias, auth, and timeout: Resolve the URL, auth params, and timeout from settings:

   URL=$(python3 "${CLAUDE_PLUGIN_ROOT}/scripts/a2a-helper.py" resolve "$URL_OR_ALIAS")
   AUTH_ARGS=()
   while IFS= read -r line; do
     AUTH_ARGS+=("$line")
   done < <(python3 "${CLAUDE_PLUGIN_ROOT}/scripts/a2a-helper.py" auth "$URL")
   SETTINGS_TIMEOUT=$(python3 "${CLAUDE_PLUGIN_ROOT}/scripts/a2a-helper.py" timeout "$URL")
   TIMEOUT_VAL="${TIMEOUT:-${SETTINGS_TIMEOUT:-120s}}"
   

3. Report effective timeout: Inform the user of the effective timeout before sending:

   Timeout: <TIMEOUT_VAL> (from: --timeout flag / settings / default 120s)

4. Build flags array:

   • If BACKGROUND: add --immediate
   • If STREAM: add --stream
   • If TIMEOUT_VAL non-empty: add --timeout "$TIMEOUT_VAL"
   • If TASK_ID: add --task "$TASK_ID"
   • If CONTEXT_ID: add --context "$CONTEXT_ID"

5. Send the message:

   a2a send "$URL" "$MESSAGE" "${FLAGS[@]}" "${AUTH_ARGS[@]}"
   

6. Track the task: Extract the task ID from the output (look for id: or "id": field). Always track the task:

   python3 "${CLAUDE_PLUGIN_ROOT}/scripts/a2a-helper.py" session add "$TASK_ID" "$URL" \
     --alias "$URL_OR_ALIAS" --message "$MESSAGE"
   

7. Handle output: Render the result based on mode (see ### Output below) and handle failures or input-required status.

Output

• Blocking (default): show the full response including artifacts. Parse and render text parts directly.
• Background: show the task ID and status. Inform the user they can check progress with the status skill.
• Streaming: relay output as it arrives.

If the task fails with a timeout error, suggest retrying with a longer value (e.g. --timeout 300s) or increasing timeout: in .claude/a2a.local.md.

If the task status is input-required, prompt the user for the required input and offer to continue the task by running send again with --task <task-id>.

Example Usage

# Send a message (blocking — waits for response)
/a2a-send https://demo.a2a-protocol.org "Summarize the latest news about AI"

# Send in background — returns task ID immediately
/a2a-send my-agent --background "Run a long analysis job"

# Continue an existing task by passing --task <task-id>
/a2a-send my-agent --task task_abc123 "Here is the additional context you asked for"