Write a brief the method can use
A Skill supplies a procedure, not your project facts. Include the information needed to apply it. When an input is missing, ask the agent to flag the gap instead of filling it with a plausible claim.
Skill: select the installed method for this task.
Audience: who will read or use the result?
Goal: what should they understand or do?
Inputs: attached files, approved facts and assets.
Output: file types, length and where to save them.
Constraints: brand rules, deadline, costs and permissions.
Missing evidence: flag it; do not invent it.Match the inputs to the output
| Output | Bring | Inspect |
|---|---|---|
| Presentation | Audience, argument, evidence and slide count | Narrative, facts, readable slides and export |
| Website | Page goals, content, brand assets and target devices | Navigation, responsive layout and working interactions |
| Image | Subject, format, reference rights and intended use | Composition, text accuracy and resolution |
| Video | Script, timing, assets and delivery format | Pacing, sound, frame quality and rendered file |
A preview shows one example. Your source material and environment can change the result. Check the Skill's own deliverables instead of expecting every package to export the same formats.
Agree on the first run
- 1
Select the method
Use the available-Skills picker or explicit invocation for the first test.
- 2
Check the plan
Confirm inputs, dependencies, output location and any external charges.
- 3
Run a small example
Start with enough content to test the procedure, without risking a large production project.
- 4
Open the files
Inspect the actual deliverable and record any mismatch with the brief.
Make the second pass specific
Separate missing facts from design preferences and execution failures. A factual error needs a corrected source; a clipped label needs a layout change; a missing render tool needs environment setup.
Keep the slide structure and approved numbers.
On slide 3, separate completed work from next steps.
Increase contrast on chart labels and remove the unsupported growth claim.
Export again and check for clipping.Record the package version, the brief and a representative output. Those three items help you evaluate future updates without depending on a chat transcript alone.
FAQ
The method may expect inputs you did not provide. Supply the missing facts or explicitly request labeled placeholders; do not treat them as finished copy.
No. Check actual dependencies, tool permissions and the output in your own environment.
Reviewed October 18, 2026 · HowAI editorial
Found something unclear? Send feedback