Skip to content

Auto-Reply Workflow

← All skills
Category Workflows Difficulty Advanced Version 1.0.0 Credits each workflow execution that calls an LLM skill consumes credits at run time; the workflow ships disabled until the user enables it Required MCP tools g8_workflow_list_node_typesg8_workflow_describe_node_typeg8_skill_create_llmg8_workflow_createg8_workflow_validateg8_workflow_update

Instead of triaging replies by hand every day, wire a durable graph8 workflow: reply lands -> LLM skill drafts a contextual follow-up -> draft is ready for review/send. Built conversationally, validated, and delivered DISABLED so the user flips it on deliberately.

Note on naming: this skill uses graph8’s g8_skill_* tools, which author LLM/API skill records that workflow Action nodes execute. Those are a different thing from Skill Library entries like this file (instructions you install into your agent). This skill composes both worlds.

When to use

  • “Automatically draft follow-ups when someone replies”
  • “Build a workflow that handles campaign replies”
  • “Stop making me triage manually every morning”

For one-off manual triage, use triage-and-reply-inbox instead.

Prerequisites

  • Activate the two tool families: g8_tool_search("workflow") and g8_tool_search("skill"), or call via g8_execute.

Steps

  1. Discover the building blocks. g8_workflow_list_node_types to see available triggers and actions; g8_workflow_describe_node_type on the reply trigger to learn its output fields (sender, subject, body, thread context). Do not guess field names; read them.
  2. Author the LLM skill. g8_skill_create_llm with a prompt template that consumes the trigger’s fields as single-brace placeholders, e.g. {sender_name}, {reply_body}. Single-brace {var} is the contract the workflow builder detects; double-brace {{var}} is treated as literal text. Use dry_run first, show the user the prompt, then persist.
  3. Design the graph. Trigger (reply received, scoped to the right campaign/mailbox) -> Action node running the LLM skill with input mappings wiring trigger outputs to the skill’s placeholders -> delivery step per the user’s preference (draft-for-review is the safe default; fully-auto send only if the user insists and understands the risk).
  4. Create disabled. g8_workflow_create with enabled=false (dry_run preview first, then persist). A workflow that has never been reviewed must not be live.
  5. Validate. g8_workflow_validate; fix any missing references or unmapped placeholders it reports. An LLM skill placeholder with no input mapping means the model gets literal {var} text at run time.
  6. Walk the user through it. Show the graph in plain language: “When X happens, Y drafts a reply using Z context, and it lands as a draft for your review.” State the per-execution credit cost.
  7. Enable on explicit approval only. g8_workflow_update with enabled=true after the user confirms. Suggest a test: send a reply to the campaign from a test address and watch the workflow produce a draft.

Example prompts

  • “When someone replies to the launch campaign, draft a follow-up in my voice for review.”
  • “Automate first-touch responses to demo requests, but let me approve sends.”
  • “Build a reply-handling workflow for the RevOps sequence.”

Troubleshooting

  • Validation reports missing references: re-check input mappings; every {placeholder} in the LLM skill needs a mapped source from the trigger.
  • Skill saved but variables not detected: you used double braces; recreate the prompt with single-brace {var}.
  • Tool not visible: both families are gated; g8_tool_search("workflow") / g8_tool_search("skill") first.

Install this skill

Add the graph8 skill marketplace once, then install auto-reply-workflow from the /plugin menu:

Terminal window
claude plugin marketplace add graph8-com/skill-library

Or install manually by saving the skill file into your project:

Terminal window
mkdir -p .claude/skills/auto-reply-workflow
curl -o .claude/skills/auto-reply-workflow/SKILL.md https://raw.githubusercontent.com/graph8-com/skill-library/v0.1.0/skills/auto-reply-workflow/SKILL.md

Connected over the graph8 MCP instead? No install needed — run g8_tool_search("library") and the same playbooks surface as callable tools.