Security Almost Never Separates a Good Agent Skill From an Average One
Security is the dimension everyone argues about. On 628 scored skills it is also the dimension that tells you the least about whether a skill is good.
We score every skill on six dimensions out of 100 points: trigger quality (20), structure (15), workflow design (25), content (10), engineering (10), security (20). Every score ships with its written rationale, and the rubric is public.
Here is the distribution of one of those dimensions across all 628 skills:
| Security score | Share of skills |
|---|---|
| Perfect 10 | 57.2% |
| 9 or above | 92.2% |
| Below 8 | 3.8% |
| Below 6 | 0.5% |
More than half the directory is flawless on security. Nine in ten are within a rounding error of it. The dimension is not measuring a spread — it's measuring a floor that almost everyone clears.
What actually does separate them
Split the directory three ways — the top 18 skills (9.5 and above), a middle band of 214 skills (7.5 to 8.5), and the 19 below 7.0 — and compare each dimension's average:
| Dimension | Top (n=18) | Middle (n=214) | Gap | Bottom (n=19) |
|---|---|---|---|---|
| Structure | 9.51 | 6.86 | 2.64 | 6.49 |
| Trigger quality | 9.42 | 7.01 | 2.41 | 4.55 |
| Engineering | 9.61 | 7.30 | 2.31 | 7.21 |
| Workflow design | 9.58 | 7.37 | 2.21 | 4.53 |
| Content | 9.67 | 7.52 | 2.15 | 6.58 |
| Security | 9.83 | 9.37 | 0.46 | 8.50 |
The gap between a top skill and a middle one is 5.7× larger on structure than it is on security.
Weight it by what the rubric actually pays
Raw gaps overstate dimensions with small weights and understate the heavy ones. Multiply each gap by the dimension's share of the score:
| Dimension | Raw gap | Weight | Contribution to the 1.97-point gap |
|---|---|---|---|
| Workflow design | 2.21 | 25% | 0.55 |
| Trigger quality | 2.41 | 20% | 0.48 |
| Structure | 2.64 | 15% | 0.40 |
| Engineering | 2.31 | 10% | 0.23 |
| Content | 2.15 | 10% | 0.21 |
| Security | 0.46 | 20% | 0.09 |
The average top skill scores 9.60; the average middle skill scores 8.00. That 1.97-point difference is what "better" means in practice — and security accounts for 0.09 of it, about 4.6%.
Workflow design, which carries the heaviest weight, is the single biggest contributor. Trigger quality is second. Structure, the dimension with the largest raw spread, is third because it's worth only 15 points.
The clearest illustration
39 skills score a perfect 10 on security while scoring below 7 on structure.
They have clean scripts and no suspicious network calls. Their SKILL.md is also a 900-line monolith with no references/ directory, no scripts/, and paths that assume the author's machine. They won't do anything malicious. They'll also struggle to do anything at all when the agent has to load them.
That's the shape of the problem: security screens for catastrophe, not for quality.
The obvious objection
"If security doesn't differentiate, why is it worth 20 points?"
Because the score is not the only thing security controls. There is a veto: confirmed credential theft, backdoors, data exfiltration, or prompt injection caps a skill at 39/100 regardless of anything else. Exactly one skill in the current directory is in that band.
So security behaves less like a ranking signal and more like a gate. Almost everything published clears it — and the things that don't are removed from consideration entirely rather than ranked low. That's a reasonable design, and it's also why the dimension looks flat: the distribution is censored at the top by design, and at the bottom by delisting.
If we wanted security to differentiate, we would have to score risk surface — how much data a skill reads, how many hosts it can reach, whether it asks for capabilities it doesn't need — instead of conformance. We're not there yet, and saying so is more useful than pretending the 10s mean more than they do.
What this means if you're choosing a skill
- Do not use the security score to choose between an 8.0 and a 9.5. At that range it is telling you almost nothing. Read the scorecard's rationale if you want the actual risk, and check the capability manifest — what it reads, what it sends, what it persists.
- Do use it as a filter. A skill in the low 7s on security, or any skill flagged in the veto band, deserves a manual look before you install it.
- Spend your attention on workflow design and trigger quality. Together they account for more than half the difference between a good skill and an average one, and both are checkable in a couple of minutes.
- Structure is the cheapest upgrade a skill author can make — and it's the dimension with the widest spread, which means the marginal skill has the most to gain there.
The data for this post comes from the same evaluated set as every scorecard on the site. The rubric and weights are at methodology; if you think a specific score is wrong, the rationale for it is published with it.