Quick wins for a faster PC:
Clear out junk files and repair common Windows errorsFree Scan →Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →Setting a pass bar before you test a Claude Code skill only pays off if the first run can actually fail. In a reported evaluation by developer Vishal Habib, the first run of a decision skill scored 0.00 because it returned a confident verdict with no evidence behind it. The fix was not a smarter answer. It was a rule that lets the skill say “can’t decide yet” and name the evidence that would settle the question.
What the author tested
Vishal Habib, in a Dev.to article published September 23, 2026, describes building three Claude Code skills for AI product managers and publishing the evaluation suite on GitHub, including the failed runs. The skill at the centre of the first failure is /build-or-not, which is meant to assess a feature idea against real examples.
![Claude AI for Beginners Bible: [5 in 1] The Ultimate Guide to Automate Your Work, Save Hours Every Week, and Use AI for Real-World Results](https://m.media-amazon.com/images/I/415+fSJacsL._SL160_.jpg)
As an Amazon Associate I earn from qualifying purchases.
The key choice in the write-up is ordering. The author says he committed the pass criteria before running any test. That matters because a bar written after the results are in can be adjusted until it fits them.
Why the first run scored 0.00
The failing test gave the skill a feature idea, no sample data, and no research tools. The skill returned “don’t build” anyway, drawing on market knowledge it recalled rather than on evidence in front of it. Under the author’s scoring, that run received 0.00.
#1 Best Overall
The root cause was a gap in the skill’s instructions. It said what to do when examples existed, but not what to do when none did. Without an instruction for that case, the skill filled the gap with the most fluent answer available, which was a verdict. A skill that always produces a decision will look productive in a demo and fail the first time the input is thin.
The fix: “no sample, no decision”
The author added a rule he calls “no sample, no decision.” It has two parts:
Rank #2
- “Can’t decide yet” becomes an accepted output when no sample is supplied, rather than a failure.
- The skill must name the specific sample that would resolve the question, so the next step is concrete.
The author reports that the next run passed the gates. This is his account of the fix and its outcome; it has not been independently reproduced.
What the eight-case comparison covered
The reported evaluation used eight cases, three runs per case, on one model. Each case was run with the skills and with plain Claude, using the same prompt. The author describes the work as a check of selected behaviours rather than a benchmark.
| Behaviour checked | Reported result with the skills |
|---|---|
| Stating a decision bar before deciding | Better than plain Claude |
| Refusing a decision without evidence | Better than plain Claude |
| Planning a rollback trigger | Better than plain Claude |
| Distinguishing a reasoned decline from an evidence gap | Better than plain Claude |
| Reporting two coverage numbers | Better than plain Claude |
| Four other cases | Plain Claude performed just as well; the cases are not itemised in the source |
The cost figure is also specific to that setup: the author reports about $2 per full run. It is not a general price for Claude Code, and other users’ runs will vary with prompts, cases, and model choice.
How much weight the numbers can bear
- All figures are the author’s own and were not independently reproduced.
- Eight cases on one model is a small sample. Results may not carry over to other models or to later Claude Code releases.
- The skill-versus-plain comparison is a behaviour check, so it shows where the skill changed what the model did, not how often real users would benefit.
- The cost applies to one author’s configuration.
Setting your own pass bar
- Write the criteria to a file and commit it before the first run. For example:
git add evals/pass-criteria.mdfollowed bygit commit -m "Pass criteria before first eval run". The commit timestamp then shows the bar predates the results. - Define a defer outcome for every decision gate. Write down what a valid “can’t decide yet” looks like, so it is not scored as a failure by default.
- Require a named sample for each decision type. If the skill cannot name one, the criteria should treat that as a pass for deferral, not a pass for a verdict.
- Run skill-on and skill-off with the same prompt, each in a fresh session. A GitHub-hosted copy of Claude Code skills documentation recommends realistic prompts with the skill enabled and disabled, and describes
claude plugin evalas running plugin-on and plugin-off cases in isolated sessions with graders. Command behaviour and setup steps change between versions, so check them against Anthropic’s current official documentation before relying on them. - Score activation and output quality separately. A skill that never loaded and a skill that loaded but answered poorly need different fixes.
- Keep the failed runs. Record them alongside the passes, with the criteria version they were scored against.
If your first run fails
- Scored zero on a no-evidence case: check whether the skill has any instruction for missing input. Add a defer rule and rerun.
- The verdict changed after you saw results: the bar moved. Restore the committed criteria file, score the existing runs against it, and only then revise.
- The skill did not appear to activate: confirm it is enabled in the session before judging its output.
- Plain Claude matches the skill on a case: record it. The author’s comparison treats that as useful information, not a reason to drop the case.
The source’s point is that a failed first run is evidence that the test works. The author’s written criteria did what they were meant to do: they exposed a skill that answered without support.
Verdict language aside, the author’s own phrasing captures the principle: “A bar set after the numbers can’t fail.”
PC Slower Than It Used to Be?
A free scan shows the junk files, broken settings and background clutter dragging Windows down - then fixes them in one click.Free scan · Windows 10 & 11Outdated Drivers Are Slowing You Down
One free scan finds every outdated or missing driver and matches the right update for your exact hardware.Free scan · exact hardware matchSource: Vishal Habib, Dev.to article, September 23, 2026. Figures are the author’s reported results and were not independently reproduced.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
