Overview
What is an Agent Kit?
An Agent Kit is a reusable, portable package for a specific AI workflow.
The idea
Teams and individuals repeat the same AI workflows every day — reviews, memos, checklists, triage, summaries. An Agent Kit packages the prompts, skills, guardrails, templates, examples, and context that make one of those workflows reusable, inspectable, and shareable.
Keep one distinction clear: OpenAI, Anthropic Claude, Google Gemini, and Ollama are AI providers — the engines that generate text. An Agent Kit is the reusable workflow package you hand to that engine. The same kit can often run against different providers.
What a kit can contain
- Prepared Prompts — reusable prompts with fill-in inputs like company name, date, goal, or output style.
- Skills — step-by-step instructions for how the AI should complete part of the job.
- Policies — guardrails for what the AI should avoid, require, or escalate.
- Templates — output formats to follow, such as a report, memo, checklist, or summary.
- Examples — sample inputs and outputs that show what good results look like.
- Context / references — reference material the AI can use while working.
A kit is distributed as a single portable .agentkit.zip file. See
Concepts & the SPEC for the package structure.
How a kit runs
When you run a kit in Forge, it collects the inputs, prepares the prompt, sends it to your selected AI provider, and lets you copy or download the result. Forge Core itself never calls AI providers and never executes scripts — it validates, renders, packages, exports, and builds AI-ready context from kits.
Agent Kits vs one-off prompts
| One-off prompt | Agent Kit |
|---|---|
| Typed once | Reusable |
| Easy to lose | Structured and versioned |
| Inconsistent | Can include inputs, templates, and guardrails |
| Hard to share | Packaged, importable, and portable across providers and tools |
Examples
- Financial Review Agent Kit — a prepared prompt for reviewing a monthly workbook, inputs for company name and reporting period, skills for formula review and trend analysis, policies requiring human review, and a client-ready summary template.
- Kubernetes SRE Agent Kit — skills for investigating pod failures, service outages, and logs; read-only policies by default; prepared prompts for incident triage; and templates for incident summaries.
- Client Memo Agent Kit — a prepared prompt that asks for client name, topic, and tone, plus a memo template, style examples, and policies for clarity and human review.
Tips
- Use Agent Kits for work you repeat — not every one-off question.
- Prepared Prompts make a kit easier to run because Forge can collect required inputs first.
- Start narrow: a kit with one clear purpose runs more consistently than a broad one.
Get Forge or read Getting started to build your first kit.