Report
Skill: pageship-report
A report tells the reader what was found, what it means, what to do about it and how sure anyone can be. The reader weighs each of those separately, so the skill never mixes them.
Your agent picks a report when the reader wants findings and a recommendation, such as the results of an investigation, a comparison of approaches, or a summary of metrics.
What the reader gets
Section titled “What the reader gets”The parts come in this order:
- Title and summary. The subject, then the main finding and the main recommendation in the source’s words, with the date and author when the source gives them.
- Observations. What was seen or measured, with where and when.
- Interpretation. What the source says the observations mean, such as a cause, a trend or a comparison. It names who made the reading when the source does.
- Recommendations. What the source says to do, with any owner and date it gives. They are numbered when the source gives an order.
- Uncertainty and open questions. What limits confidence: missing data, small samples, unverified claims, sources that disagree. Every gap in the source becomes an open question here.
- Sources. As the source names them.
What the skill works from
Section titled “What the skill works from”The source is your request and the files and data it points to. The agent reads them and the project’s glossary, then sorts every statement into one of the four parts: observation, interpretation, recommendation or uncertainty.
It adds no finding, cause or recommendation of its own unless you ask for its analysis. When you do, it labels each one as its own. Every number keeps the label the source gives it, such as measured, estimate or target.
Patterns it often uses
Section titled “Patterns it often uses”Reports often show numbers, so they tend to use:
- Key figures for the few headline numbers after the summary.
- Charts for a trend, a ranking or a spread.
- Tables for exact values to look up or compare.
Example request
Section titled “Example request”Write a report on last month's support tickets from tickets-september.csv.
Compare response times by queue and say what we should change. Publish it.