Claude Skill and Prompt Tune-up

Outcomes

• A one-page diagnosis of what causes the weak results, such as an unclear job, no output format, no decision rules, no checks, or a skill description that does not say when it applies
• The improved version: a SKILL.md with YAML front matter (name and description) and instructions, or a rewritten prompt, plus a paste-in version for any Claude chat
• A test log of 3 rounds on your sample inputs. Round 1 runs the original, rounds 2 and 3 follow our changes, with every output and a score against the criteria you gave us
• A one-page review checklist: what a person must check in the output before using it

What we do

• Read what you send and your idea of a good output, and ask up to 5 questions by email if something is unclear
• Rewrite it in plain, specific instructions. We change only what the diagnosis supports, and we say what we changed and why
• Run all your samples in Claude, in 3 rounds, and keep every output in the log
• Score each round against up to 5 criteria we agree with you, change the instructions after each round, and show whether the result got better, worse or stayed the same
• Deliver the files by email and answer one round of questions within 7 days

What you provide

• The prompt or skill, up to 1,500 words, and any reference files a skill uses (up to 3 files, 2,000 words in all)
• Up to 3 sample inputs, each up to 2 pages, with personal and customer data removed, and an example of an output you consider good
• Up to 5 criteria for a good output, in plain words (for example: under 150 words, lists the three next steps, never quotes a price)
• A reviewer on your side who can judge quality and reply within 2 working days

Timeline 5 working days from the day we have the form and your samples. Working days are counted from the day we hold your completed intake form and the items under "What you provide". The clock pauses while we wait for you.

Limits

• One prompt or skill, 3 test rounds, up to 3 sample inputs and up to 5 criteria. English only, unless you supply Arabic samples and can judge the result
• We test in a Claude chat that we have access to when we do the work. Results can differ on your plan, on another model or version, and from run to run
• Text in, text out. Nothing connects to Zeyto or any other system
• Your samples are used only for this job

Not included

• Building a new skill from scratch (see the Custom Claude Skills and Prompt Workflows Build package)
• A guarantee that outputs are correct. Claude can be wrong. Accounting, tax, legal and HR outputs must be checked by a qualified person
• Any automation that sends, posts, pays or logs in by itself. A person reviews and sends everything
• A Claude subscription, or training for your team

Support after delivery

• One round of questions by email within 7 days
• If the improved version gives a poor result on an input of the same kind as your samples within 14 days, send it and we adjust once
• Refunds follow the Zeyto Marketplace Terms (section 6): write to support@zeyto.app within 7 days of purchase if the service was not delivered or differs materially from this description.