/plan We just finished [the task we just did]. Turn it into a reusable skill at .grok/skills/[skill-name]/SKILL.md. Include: a description of when to use it, [inputs and outputs], the exact steps we ended up using (not the ones that failed), the gotchas we hit ([gotchas]) as checks, and a final step that shows me row counts before and after and stops. If we wrote a script, save it inside the skill folder and call it from the steps. Set disable-model-invocation: true.
Turn a finished task into a documented skill
You and the agent just figured out a fiddly job together, and you never want to figure it out from scratch again.
- Grok Build
- Claude Code
For: Tech person, Owner, Office manager · Updated October 25, 2026
A starting point: fill in the [brackets] with your own details, and check the answer before you use it. Tool screens change often, so yours may look a little different.
Video walkthrough
Video coming later: Teach Claude Your Process Once: Skills in 3 Minutes (P2-09). Until then, watch the free episodes that are out now.
Also covered in: P3-05 · Which AI Coding Agent for a Cost-Conscious Small Business? The Part 3 Verdict (Part 3: Coding Agents Compared) · coming later
The prompts
Copy the version for the tool you use, then replace every [bracket].
Turn what we just did ([the task we just did]) into a project skill at .claude/skills/[skill-name]/SKILL.md: when to use it, [inputs and outputs], the working steps, our [gotchas] as checks, and a final before/after summary. Bundle any script we wrote in the skill folder. Show me before saving.
Fill in these [brackets]
| Bracket | What to put there |
|---|---|
[the task we just did] | e.g. cleaned the monthly invoice export so it imports into our bookkeeping app |
[skill-name] | e.g. clean-invoice-export |
[inputs and outputs] | e.g. input: exports/invoices-YYYY-MM.csv; output: clean/invoices-YYYY-MM.csv |
[gotchas] | What went wrong the first time. e.g. dates came as DD/MM; refunds were negative; one header had a trailing space |
Docs note (checked Oct 17, 2026): Grok Build discovers skills in ./.grok/skills/ (walked up to the repo root) and ~/.grok/skills/, shows user-invocable skills as slash commands, and also reads Claude Code skills. A skill is a folder of Markdown instructions plus any scripts and resources. grok inspect lists the skills, rules, and plugins it found. One difference to know: in Grok Build, allowed-tools in a skill does not grant or restrict tools; in Claude Code it pre-approves the listed tools for that turn.
What a good answer includes (our checklist, not an AI answer):
- The steps that finally worked, not the first attempt
- Each problem you hit turned into a check or a rule
- Inputs and outputs named clearly
- A stop before anything is committed, published, or sent
- Shown to you before it’s saved
This one needs a real coding-agent session in your own project, so we didn’t run it in a web chat on Oct 25, 2026.
Check before you send
- Steps are the ones that worked, not the first attempt
- Every gotcha became a check the skill runs
- It ends with numbers you can verify (row counts, totals)
- Confirm it loads: grok inspect (Grok Build) or /skills
- Run it next month on a copy of the file first
Privacy note: what not to paste
Use a sample or redacted export while building the skill. Never store real customer data or bank details inside the skill folder.