How this helps you
You get a repeatable speed audit with real before and after numbers instead of guesses, and a branch you can review rather than changes you have to trust.
Save the skill at https://raw.githubusercontent.com/sebbsssss/optimize-skill/main/skills/optimize/SKILL.md as a skill called optimize. Then, in the repo [repo folder], run the optimize audit, audit only. Do not change any code or commit. When you finish, show me the baseline numbers, the verdict for each item, and anything you could not measure and why.
Fill in: [repo folder]: the folder your Bot cloned. Say 'frontend', 'backend' or 'database' to limit the scope.
What they did
@sebbsssss shared /optimize, a skill he actively uses with his Grok Bot. It checks a web app for speed problems such as N+1 database queries, repeated reads, serverless connections, sequential API calls, heavy JavaScript and slow waits for AI answers. The skill measures a baseline first, applies safe fixes on a separate branch, then runs the same test again and lists what it could not measure. On a simple internal app he reports it found a duplicate page load and repeated data reads, and that fixing them cut the time from choosing a role to seeing the overview by 32 to 40 percent on repeat visits (his app, not a promise for yours). The repo is free and MIT licensed. Its README says there is no verified import button for Grok Bot: paste the raw SKILL.md link and ask the Bot to save it.
Who it's for: People who build or maintain a web app with a coding Bot and want safe, measured speed fixes.
Before you start
- A web app in a git repo that your Bot's computer can build and run
- A Bot with shell access to that repo (see the tip on running a coding CLI on your Bot's computer)
- A dev database only if you want query timings
How to try it
- Open a Bot that already has your repo. Paste the prompt below with the raw SKILL.md link and ask it to save the skill as optimize.
- Ask for audit only first. The skill's modes include 'audit only' or 'report only', which changes no code.
- Read the report. Each of the 31 items gets one verdict (Pass, Fixed, Recommend, Needs decision or N/A) with evidence, and a measurement script is saved in the repo.
- If you agree, ask for the full run. It applies only safe fixes on a new branch, re-runs the same script and writes OPTIMIZE_REPORT.md with before and after numbers.
- Review the branch diff yourself, run your own tests, and commit or merge it yourself. The skill does not commit, push or deploy.
- Work through the Recommend and Needs decision items as separate jobs, since indexes, caching layers, queues and API changes need your approval.
What to expect
The author's example: on a simple internal app the audit found a duplicate page load and repeated data reads. After fixes the time from picking a role to seeing the overview fell by 32 to 40 percent on repeat visits. Your numbers will differ, and the skill is told to write 'not measured' where it cannot measure.
Common mistakes
- Running the full mode on production or without reading the audit first.
- Expecting it to install Redis, queues, a CDN or paid tools. It only recommends those.
- Approving API changes such as adding pagination without checking who uses the endpoint.
- Comparing numbers from different machines. The skill runs one script before and after on the same setup; keep it that way.
- Merging without your own tests.
Related tips
- Run Claude Code or Codex on your Bot's computer with your own plan: Your Bot's computer has a terminal, so it can install a coding CLI, you sign in once, and your Bots can use that subscription from anywhere.
- Make your Bot the manager, not the coder: one rule and locked doors: Let your Bot plan and check the work while coding agents write the code. Start with one rule, and remember every Bot shares one computer and its logins.
- What actually stops a coding Bot (and what is only advice): Of 15 operations tested, only a merge to main hit a hard stop. Skill-file rules are only advice.
Worth knowingThe value is in the rule set: measure first, branch, same test after, list what you could not measure.Next in Skills →Save a skill that runs whenever you say a trigger phrase
Learn more: How to Use Grok Bot: A Beginner's Guide (2026) · All Grok Bot skills tips