Orange Ink · Practical agent patterns for builders
Evaluate the loop — not just the final paragraph
A polished answer can hide a wrong source, unnecessary tool call, or missed escalation. Score the path as well as the output.
- ✓Normal, ambiguous, stale, empty, and failure cases
- ✓Evidence quality and tool selection scoring
- ✓Stop-rule and human-handoff checks
What you get
Instant download · Open beside your tools · Keep forever
Before you add another agent framework…
Are you calling a prompt chain an agent without defining what it may do?
Do demos look clever while real tool calls fail on missing context?
Are you unsure where chat ends and an agent becomes worthwhile?
Could you explain how you will test the agent before it touches live work?
Free pack starter
Free download
Open the files
Build the loop, not the hype
Use the PDFs beside your editor, automation tool, or agent platform. The patterns are vendor-neutral and deliberately small enough to test.
Agent-or-chat decision
A practical test for deciding whether a task needs tools and iteration, or simply a well-briefed chat.
Brief → tool → review loop
Write the goal, evidence, permissions, stop conditions, and handoff before implementation.
Three starter skills
Research brief, inbox triage, and content QA patterns with inputs, tool contracts, and review gates.
Evaluation scorecard
Test normal, ambiguous, stale, and failure cases before the agent reaches live work.
Simple path
How to use the kit
- 1
Run pre-flight
Choose a repetitive, reversible task and write down what the agent must never do.
- 2
Build one narrow loop
Give it the minimum tools, explicit evidence requirements, and a human approval gate.
- 3
Evaluate before expanding
Score a small test set, inspect failures, then decide whether more autonomy is earned.
Get the pack
The complete pack
AI Agents Starter Kit
An 18-page practical mini-course for choosing the right agent job, writing a tool-ready brief, building the brief → tool → review loop, evaluating outputs, and installing three reusable starter skills.
Vendor-neutral agent design
Works whether you code, use an automation platform, or configure an agent builder.
Three reusable skill patterns
Concrete research, triage, and QA blueprints — distinct from generic business prompt libraries.
Evaluation and failure recovery
Define evidence, confidence, retries, escalation, and human acceptance before live use.
Human review by design
No money bots, unsupervised sending, or autonomy theatre. The operator remains accountable.
7-day goodwill fit guarantee
If the kit is clearly not a fit, email support@orangeink.consulting about an eligible refund.
Secure checkout
Get the pack
One payment. Instant PDF. No subscription.
£97.00Today £47.00
Soft-launch pricing
What you download
Questions, answered
Before you decide
Who is this for?
Builders, freelancers, and automation-curious operators who want to design a small useful agent. You do not need to be an AI researcher, but you should be willing to test outputs and own the final action.
How is this different from the ChatGPT business kit?
This is about agent architecture: tool permissions, iterative loops, stop conditions, evaluation, and reusable skills. It is not a weekly ChatGPT operations prompt system.
Do I need to code?
No. The patterns work in code and no-code agent builders. The kit focuses on the specification and evaluation decisions every implementation needs.
Does this teach autonomous money-making agents?
No. It explicitly rejects set-and-forget money bots. Start with reversible work, keep external actions behind approval, and expand only when evidence supports it.
What do I receive?
The free Pre-Flight Checklist is an 8-page PDF. The paid Starter Kit is an 18-page PDF mini-course. The optional Prompt Pack and 5-page Build Workbook are separate checkout add-ons.
Is this professional advice?
No. It is educational operational guidance. Do not use agents as substitutes for legal, financial, medical, safety, or other regulated judgement.
Next step
Start with the free go/no-go check
Use the 8-page checklist to scope one candidate workflow. Move to the Starter Kit when you are ready to specify, test, and operate it.