Retour à tous les articles
ArticleEnglish

Build Your First Agent Skill for a Weekly Report

Auteur:

Turn recurring reporting rules into a small Agent Skill while keeping each week's facts separate. Use a complete SKILL.md and check the draft against its source notes.

An open lime folder holding one blank instruction sheet and a separate stack of weekly note cards

You have a prompt that produces a usable weekly report. The next Friday, you paste the same rules again: separate completed work from plans, show what is blocked, and do not invent dates. The notes change; the rules do not. This is a reasonable point to consider an Agent Skill—if the assistant you use supports Skills.

A Skill is a folder with a SKILL.md file. Its description helps a supporting agent decide when to use it; its body gives the recurring instructions. The Agent Skills specification defines the file format and optional resources. It does not guarantee that every app will discover a folder you create or invoke it for every request. Check your chosen app's installation instructions before you rely on the file.

Separate the rule from this week's facts

Suppose these are your notes for a fictional week:

Mon: Finished the event budget draft.
Tue: Sent the budget to Lee for review; no reply recorded.
Thu: Reproduced a registration-form error; the fix is not started.
Fri: Plan to invite participants after approval. No date agreed.

The three-section format and the rule against invented approvals belong in the Skill. Lee's review status and the registration error belong in the current request. If you store those facts in the Skill, next week's report may repeat them after they have changed.

Create a folder named weekly-report in the location your supporting app documents. Put this file inside it:

---
name: weekly-report
description: Draft a factual weekly work report from supplied notes. Use when a user asks to turn work updates into a weekly status report.
---

# Weekly report

Read the notes supplied for this report. If there are no notes, ask for them.

Return three sections: Completed, Waiting, and Next work.
Keep finished work separate from plans and from work that is only in progress.
Do not invent a deadline, approval, owner, or result.
Mark an important missing fact as not confirmed.
Before returning the draft, compare each statement with the supplied notes.
Return the report for review. Do not send or publish it.

The folder name and name match the format's naming rule. description says both what the Skill does and when it should be used. The instructions are short because this job has one repeated output and one important failure mode. Add a template or reference file only when a real task needs one.

Give it current input and inspect the result

After installing the folder according to your app's instructions, make a request such as: “Use the weekly-report Skill to draft this week's report from the notes below. Return a draft only.” Paste the four notes. Explicitly naming the Skill is a useful first check of whether the app can access it; later you can try an ordinary request to see whether it selects the Skill without that cue. Neither check proves a general trigger rate.

A defensible draft would place the event budget draft under Completed. It would say that review is pending under Waiting. It could place the planned invitations and form fix under Next work, while retaining that there is no agreed invitation date and no fix yet. “Lee approved the budget” or “the registration form was fixed” would fail even if the report sounded polished.

Try one more request with no notes. The Skill says to ask for the missing input. If the assistant writes a plausible weekly report anyway, you have learned something important about this setup. Correct the instructions or the way you invoke them before using it on a real update.

Keep a plain saved prompt if this is an occasional task or your app cannot use Skills. The value of the package is that you can maintain the recurring rules in one place; the report still requires new facts and a human check. For a fuller comparison of the two approaches, see Agent Skills vs Prompts.