Cisco AI Defense skill scanner: evaluation results

What changes when the LLM judge and the System One model are switched on.

NVIDIA SkillSpector vs Cisco skill-scanner

Both scanners run over one population of skill directories, byte-identical on each side, with every analyzer each tool ships enabled and neither tuned on these labels. Three arms: static analysis only, then both tools on Gemma 4 26B, then both on Claude Haiku 4.5.

The comparison is possible because SkillSpector emits the same LOW/MEDIUM/HIGH/CRITICAL levels this scanner does, so a threshold means the same thing on both sides. Package-level decisions do not translate, and are reported separately rather than normalised into a shared verdict neither tool emits.

Only records where both tools returned a usable result are scored. A crashed scan, an unparsable payload and a run with analyzers switched off all present as zero findings, so scoring them as clean would credit specificity that was never earned.

What was compared

VersionConfiguration
NVIDIA SkillSpectorv2.11.2 (94f5cc80679c45a1fa4774e2d37ad663ca0be337)All 27 analyzers enabled; built-in YARA; live OSV; baselines off; resource ceilings raised; --transitive off (corpus safety terms)
Cisco skill-scanner4f8bec3Shipped core rule pack; judge + meta-judge; escalate-only verdict repair; policy suppressions empty; VirusTotal and AI Defense off (no credential)

Four decision lenses

This scanner emits findings plus a package verdict. SkillSpector emits findings plus a 0–100 risk score, banded into a recommendation, and gates an install above 50. Its score accumulates with diminishing returns per rule, so a single CRITICAL finding at 0.95 confidence scores 47 and returns CAUTION, which does not gate.

A gate-to-gate comparison therefore measures two different thresholds rather than two engines. Detection at a fixed severity is the primary lens for that reason; each tool's own gate is reported alongside it, and a sweep across every threshold is given below so the operating point is visible rather than assumed.

Each lens carries a paired difference: both tools are resampled together on the same records, so shared corpus difficulty cancels rather than widening both intervals, and an interval excluding zero means the two tools are separable on that lens. Each corpus also reports complementarity, the share of records exactly one tool flags, which is the interesting population because neither tool is a superset of the other.

Same model: Gemma 4 26B

Both tools on bedrock-mantle/google.gemma-4-26b-a4b.

MaliciousSkillBench, source-disjoint split

1367 records scored on both tools · 528 benign, 839 malicious

Any finding at MEDIUM or above

ToolRecallPrecisionF1False-positive rate
Cisco skill-scanner (Gemma 4 26B)48.99%81.55%61.21%17.61%
NVIDIA SkillSpector (Gemma 4 26B)79.98%65.66%72.11%66.48%

Paired recall difference (Cisco skill-scanner (Gemma 4 26B) minus NVIDIA SkillSpector (Gemma 4 26B)): -31.0 points, 95% interval [-34.2, -27.5] — statistically separable.

Any finding at HIGH or above

ToolRecallPrecisionF1False-positive rate
Cisco skill-scanner (Gemma 4 26B)25.27%80.00%38.41%10.04%
NVIDIA SkillSpector (Gemma 4 26B)75.45%69.41%72.30%52.84%

Paired recall difference (Cisco skill-scanner (Gemma 4 26B) minus NVIDIA SkillSpector (Gemma 4 26B)): -50.2 points, 95% interval [-53.9, -46.5] — statistically separable.

Each tool's own shipped install gate

ToolRecallPrecisionF1False-positive rate
Cisco skill-scanner (Gemma 4 26B)25.27%80.00%38.41%10.04%
NVIDIA SkillSpector (Gemma 4 26B)38.86%74.43%51.06%21.21%

Cisco skill-scanner (Gemma 4 26B) matches the HIGH row exactly wherever that happens below. It is a property of the default policy, which gates at HIGH, not a repeated table: the decision was read from each tool's emitted verdict field on every record.

Paired recall difference (Cisco skill-scanner (Gemma 4 26B) minus NVIDIA SkillSpector (Gemma 4 26B)): -13.6 points, 95% interval [-16.3, -10.7] — statistically separable.

Block on any finding at all

ToolRecallPrecisionF1False-positive rate
Cisco skill-scanner (Gemma 4 26B)92.61%61.86%74.18%90.72%
NVIDIA SkillSpector (Gemma 4 26B)80.93%64.42%71.74%71.02%

Paired recall difference (Cisco skill-scanner (Gemma 4 26B) minus NVIDIA SkillSpector (Gemma 4 26B)): +11.7 points, 95% interval [+8.7, +14.9] — statistically separable.

Where they disagree

Both flaggedNeither flaggedOnly Cisco skill-scanner (Gemma 4 26B)Only NVIDIA SkillSpector (Gemma 4 26B)AgreementComplementarity
4803212454258.60%41.40%

Cost of a scan

ToolMean seconds per skillInput tokensOutput tokens
Cisco skill-scanner (Gemma 4 26B)not instrumentednot instrumentednot instrumented
NVIDIA SkillSpector (Gemma 4 26B)26.018not instrumentednot instrumented

Wall clock was recorded per subprocess, which covers SkillSpector but not this scanner's in-process path, and neither tool reported token usage through the harness. The uninstrumented cells are marked rather than shown as zero, and no cost comparison is drawn from this table. Judge token cost was measured separately and is reported in the token-cost section.

Integrity

ToolRowsUsableErrorsCapability degradedReported partial
Cisco skill-scanner (Gemma 4 26B)13841384000
NVIDIA SkillSpector (Gemma 4 26B)138413670171043

MaliciousSkillBench, balanced subset

793 records scored on both tools · 397 benign, 396 malicious

Any finding at MEDIUM or above

ToolRecallPrecisionF1False-positive rate
Cisco skill-scanner (Gemma 4 26B)82.07%78.50%80.25%22.42%
NVIDIA SkillSpector (Gemma 4 26B)94.44%57.54%71.51%69.52%

Paired recall difference (Cisco skill-scanner (Gemma 4 26B) minus NVIDIA SkillSpector (Gemma 4 26B)): -12.4 points, 95% interval [-16.2, -8.7] — statistically separable.

Any finding at HIGH or above

ToolRecallPrecisionF1False-positive rate
Cisco skill-scanner (Gemma 4 26B)65.91%93.21%77.22%4.79%
NVIDIA SkillSpector (Gemma 4 26B)86.36%63.22%73.00%50.13%

Paired recall difference (Cisco skill-scanner (Gemma 4 26B) minus NVIDIA SkillSpector (Gemma 4 26B)): -20.5 points, 95% interval [-25.0, -15.8] — statistically separable.

Each tool's own shipped install gate

ToolRecallPrecisionF1False-positive rate
Cisco skill-scanner (Gemma 4 26B)65.91%93.21%77.22%4.79%
NVIDIA SkillSpector (Gemma 4 26B)70.96%80.52%75.44%17.13%

Paired recall difference (Cisco skill-scanner (Gemma 4 26B) minus NVIDIA SkillSpector (Gemma 4 26B)): -5.1 points, 95% interval [-9.1, -1.0] — statistically separable.

Block on any finding at all

ToolRecallPrecisionF1False-positive rate
Cisco skill-scanner (Gemma 4 26B)96.97%50.73%66.61%93.95%
NVIDIA SkillSpector (Gemma 4 26B)95.20%54.56%69.37%79.09%

Paired recall difference (Cisco skill-scanner (Gemma 4 26B) minus NVIDIA SkillSpector (Gemma 4 26B)): +1.8 points, 95% interval [-0.3, +4.3] — not separable from zero.

Where they disagree

Both flaggedNeither flaggedOnly Cisco skill-scanner (Gemma 4 26B)Only NVIDIA SkillSpector (Gemma 4 26B)AgreementComplementarity
3971261725365.95%34.05%

Cost of a scan

ToolMean seconds per skillInput tokensOutput tokens
Cisco skill-scanner (Gemma 4 26B)not instrumentednot instrumentednot instrumented
NVIDIA SkillSpector (Gemma 4 26B)23.174not instrumentednot instrumented

Integrity

ToolRowsUsableErrorsCapability degradedReported partial
Cisco skill-scanner (Gemma 4 26B)800800000
NVIDIA SkillSpector (Gemma 4 26B)80079307674

OpenSkillRisk

343 records scored on both tools · 188 contextually_risky, 155 obviously_malicious

Any finding at MEDIUM or above

ToolRecallPrecisionF1False-positive rate
Cisco skill-scanner (Gemma 4 26B)85.71%no harmless classno harmless classno harmless class
NVIDIA SkillSpector (Gemma 4 26B)95.92%no harmless classno harmless classno harmless class

Paired recall difference (Cisco skill-scanner (Gemma 4 26B) minus NVIDIA SkillSpector (Gemma 4 26B)): -10.2 points, 95% interval [-14.0, -6.4] — statistically separable.

Any finding at HIGH or above

ToolRecallPrecisionF1False-positive rate
Cisco skill-scanner (Gemma 4 26B)71.43%no harmless classno harmless classno harmless class
NVIDIA SkillSpector (Gemma 4 26B)86.30%no harmless classno harmless classno harmless class

Paired recall difference (Cisco skill-scanner (Gemma 4 26B) minus NVIDIA SkillSpector (Gemma 4 26B)): -14.9 points, 95% interval [-19.0, -10.8] — statistically separable.

Each tool's own shipped install gate

ToolRecallPrecisionF1False-positive rate
Cisco skill-scanner (Gemma 4 26B)71.43%no harmless classno harmless classno harmless class
NVIDIA SkillSpector (Gemma 4 26B)77.84%no harmless classno harmless classno harmless class

Paired recall difference (Cisco skill-scanner (Gemma 4 26B) minus NVIDIA SkillSpector (Gemma 4 26B)): -6.4 points, 95% interval [-10.5, -2.3] — statistically separable.

Block on any finding at all

ToolRecallPrecisionF1False-positive rate
Cisco skill-scanner (Gemma 4 26B)99.42%no harmless classno harmless classno harmless class
NVIDIA SkillSpector (Gemma 4 26B)98.25%no harmless classno harmless classno harmless class

Paired recall difference (Cisco skill-scanner (Gemma 4 26B) minus NVIDIA SkillSpector (Gemma 4 26B)): +1.2 points, 95% interval [-0.3, +2.9] — not separable from zero.

Where they disagree

Both flaggedNeither flaggedOnly Cisco skill-scanner (Gemma 4 26B)Only NVIDIA SkillSpector (Gemma 4 26B)AgreementComplementarity
289954086.88%13.12%

Cost of a scan

ToolMean seconds per skillInput tokensOutput tokens
Cisco skill-scanner (Gemma 4 26B)not instrumentednot instrumentednot instrumented
NVIDIA SkillSpector (Gemma 4 26B)23.519not instrumentednot instrumented

Integrity

ToolRowsUsableErrorsCapability degradedReported partial
Cisco skill-scanner (Gemma 4 26B)374374000
NVIDIA SkillSpector (Gemma 4 26B)374343031302

HarmfulSkillBench

196 records scored on both tools · 196 unlabelled

Any finding at MEDIUM or above

ToolRecallPrecisionF1False-positive rate
Cisco skill-scanner (Gemma 4 26B)56.12%not permittednot permittednot permitted
NVIDIA SkillSpector (Gemma 4 26B)84.18%not permittednot permittednot permitted

Paired recall difference (Cisco skill-scanner (Gemma 4 26B) minus NVIDIA SkillSpector (Gemma 4 26B)): -28.1 points, 95% interval [-35.7, -20.7] — statistically separable.

Any finding at HIGH or above

ToolRecallPrecisionF1False-positive rate
Cisco skill-scanner (Gemma 4 26B)38.78%not permittednot permittednot permitted
NVIDIA SkillSpector (Gemma 4 26B)57.65%not permittednot permittednot permitted

Paired recall difference (Cisco skill-scanner (Gemma 4 26B) minus NVIDIA SkillSpector (Gemma 4 26B)): -18.9 points, 95% interval [-26.5, -11.2] — statistically separable.

Each tool's own shipped install gate

ToolRecallPrecisionF1False-positive rate
Cisco skill-scanner (Gemma 4 26B)38.78%not permittednot permittednot permitted
NVIDIA SkillSpector (Gemma 4 26B)29.08%not permittednot permittednot permitted

Paired recall difference (Cisco skill-scanner (Gemma 4 26B) minus NVIDIA SkillSpector (Gemma 4 26B)): +9.7 points, 95% interval [+2.0, +17.3] — statistically separable.

Block on any finding at all

ToolRecallPrecisionF1False-positive rate
Cisco skill-scanner (Gemma 4 26B)98.47%not permittednot permittednot permitted
NVIDIA SkillSpector (Gemma 4 26B)87.76%not permittednot permittednot permitted

Paired recall difference (Cisco skill-scanner (Gemma 4 26B) minus NVIDIA SkillSpector (Gemma 4 26B)): +10.7 points, 95% interval [+6.6, +15.3] — statistically separable.

Where they disagree

Both flaggedNeither flaggedOnly Cisco skill-scanner (Gemma 4 26B)Only NVIDIA SkillSpector (Gemma 4 26B)AgreementComplementarity
1012296462.76%37.24%

Cost of a scan

ToolMean seconds per skillInput tokensOutput tokens
Cisco skill-scanner (Gemma 4 26B)not instrumentednot instrumentednot instrumented
NVIDIA SkillSpector (Gemma 4 26B)17.079not instrumentednot instrumented

Integrity

ToolRowsUsableErrorsCapability degradedReported partial
Cisco skill-scanner (Gemma 4 26B)200200000
NVIDIA SkillSpector (Gemma 4 26B)20019604161

Same model: Claude Haiku 4.5

Both tools on us.anthropic.claude-haiku-4-5 via Bedrock.

MaliciousSkillBench, source-disjoint split

1322 records scored on both tools · 518 benign, 804 malicious

Any finding at MEDIUM or above

ToolRecallPrecisionF1False-positive rate
Cisco skill-scanner (core rules)45.77%81.96%58.74%15.64%
NVIDIA SkillSpector84.58%64.09%72.92%73.55%

Paired recall difference (Cisco skill-scanner (core rules) minus NVIDIA SkillSpector): -38.8 points, 95% interval [-42.7, -35.2] — statistically separable.

Any finding at HIGH or above

ToolRecallPrecisionF1False-positive rate
Cisco skill-scanner (core rules)26.87%86.06%40.95%6.76%
NVIDIA SkillSpector74.50%67.45%70.80%55.79%

Paired recall difference (Cisco skill-scanner (core rules) minus NVIDIA SkillSpector): -47.6 points, 95% interval [-51.4, -43.9] — statistically separable.

Each tool's own shipped install gate

ToolRecallPrecisionF1False-positive rate
Cisco skill-scanner (core rules)26.87%86.06%40.95%6.76%
NVIDIA SkillSpector54.85%74.62%63.23%28.96%

Paired recall difference (Cisco skill-scanner (core rules) minus NVIDIA SkillSpector): -28.0 points, 95% interval [-31.7, -24.4] — statistically separable.

Block on any finding at all

ToolRecallPrecisionF1False-positive rate
Cisco skill-scanner (core rules)74.63%57.42%64.90%85.91%
NVIDIA SkillSpector84.58%64.09%72.92%73.55%

Paired recall difference (Cisco skill-scanner (core rules) minus NVIDIA SkillSpector): -10.0 points, 95% interval [-13.8, -6.2] — statistically separable.

Where they disagree

Both flaggedNeither flaggedOnly Cisco skill-scanner (core rules)Only NVIDIA SkillSpectorAgreementComplementarity
4202322964149.32%50.68%

Cost of a scan

ToolMean seconds per skillInput tokensOutput tokens
Cisco skill-scanner (core rules)13.76313,367,9822,107,191
NVIDIA SkillSpector39.585not instrumentednot instrumented

Integrity

ToolRowsUsableErrorsCapability degradedReported partial
Cisco skill-scanner (core rules)1384135103344
NVIDIA SkillSpector138413550291045

MaliciousSkillBench, balanced subset

766 records scored on both tools · 388 benign, 378 malicious

Any finding at MEDIUM or above

ToolRecallPrecisionF1False-positive rate
Cisco skill-scanner (core rules)80.69%73.85%77.12%27.84%
NVIDIA SkillSpector97.88%53.94%69.55%81.44%

Paired recall difference (Cisco skill-scanner (core rules) minus NVIDIA SkillSpector): -17.2 points, 95% interval [-21.2, -13.4] — statistically separable.

Any finding at HIGH or above

ToolRecallPrecisionF1False-positive rate
Cisco skill-scanner (core rules)63.23%89.18%73.99%7.47%
NVIDIA SkillSpector89.42%61.34%72.77%54.90%

Paired recall difference (Cisco skill-scanner (core rules) minus NVIDIA SkillSpector): -26.2 points, 95% interval [-31.0, -21.3] — statistically separable.

Each tool's own shipped install gate

ToolRecallPrecisionF1False-positive rate
Cisco skill-scanner (core rules)63.23%89.18%73.99%7.47%
NVIDIA SkillSpector81.48%73.68%77.39%28.35%

Paired recall difference (Cisco skill-scanner (core rules) minus NVIDIA SkillSpector): -18.3 points, 95% interval [-22.8, -13.5] — statistically separable.

Block on any finding at all

ToolRecallPrecisionF1False-positive rate
Cisco skill-scanner (core rules)92.86%50.36%65.30%89.18%
NVIDIA SkillSpector97.88%53.86%69.48%81.70%

Paired recall difference (Cisco skill-scanner (core rules) minus NVIDIA SkillSpector): -5.0 points, 95% interval [-7.4, -2.4] — statistically separable.

Where they disagree

Both flaggedNeither flaggedOnly Cisco skill-scanner (core rules)Only NVIDIA SkillSpectorAgreementComplementarity
40774627962.79%37.21%

Cost of a scan

ToolMean seconds per skillInput tokensOutput tokens
Cisco skill-scanner (core rules)19.8658,650,4651,814,253
NVIDIA SkillSpector30.982not instrumentednot instrumented

Integrity

ToolRowsUsableErrorsCapability degradedReported partial
Cisco skill-scanner (core rules)80078301734
NVIDIA SkillSpector800783017674

Static only, no model

Deterministic analyzers on both sides. No LLM.

MaliciousSkillBench, source-disjoint split

1368 records scored on both tools · 529 benign, 839 malicious

Any finding at MEDIUM or above

ToolRecallPrecisionF1False-positive rate
Cisco skill-scanner (core rules)7.99%60.36%14.11%8.32%
NVIDIA SkillSpector34.33%56.14%42.60%42.53%

Paired recall difference (Cisco skill-scanner (core rules) minus NVIDIA SkillSpector): -26.3 points, 95% interval [-29.8, -23.2] — statistically separable.

Any finding at HIGH or above

ToolRecallPrecisionF1False-positive rate
Cisco skill-scanner (core rules)7.75%61.90%13.77%7.56%
NVIDIA SkillSpector14.54%60.10%23.42%15.31%

Paired recall difference (Cisco skill-scanner (core rules) minus NVIDIA SkillSpector): -6.8 points, 95% interval [-9.3, -4.1] — statistically separable.

Each tool's own shipped install gate

ToolRecallPrecisionF1False-positive rate
Cisco skill-scanner (core rules)7.75%61.90%13.77%7.56%
NVIDIA SkillSpector1.31%57.89%2.56%1.51%

Paired recall difference (Cisco skill-scanner (core rules) minus NVIDIA SkillSpector): +6.4 points, 95% interval [+4.8, +8.3] — statistically separable.

Block on any finding at all

ToolRecallPrecisionF1False-positive rate
Cisco skill-scanner (core rules)60.55%52.48%56.23%86.96%
NVIDIA SkillSpector34.33%55.38%42.38%43.86%

Paired recall difference (Cisco skill-scanner (core rules) minus NVIDIA SkillSpector): +26.2 points, 95% interval [+21.6, +31.0] — statistically separable.

Where they disagree

Both flaggedNeither flaggedOnly Cisco skill-scanner (core rules)Only NVIDIA SkillSpectorAgreementComplementarity
818253043266.23%33.77%

Cost of a scan

ToolMean seconds per skillInput tokensOutput tokens
Cisco skill-scanner (core rules)0.015not instrumentednot instrumented
NVIDIA SkillSpector2.453not instrumentednot instrumented

Integrity

ToolRowsUsableErrorsCapability degradedReported partial
Cisco skill-scanner (core rules)138413840011
NVIDIA SkillSpector138413680161043

MaliciousSkillBench, balanced subset

794 records scored on both tools · 397 benign, 397 malicious

Any finding at MEDIUM or above

ToolRecallPrecisionF1False-positive rate
Cisco skill-scanner (core rules)33.00%94.24%48.88%2.02%
Cisco skill-scanner (all rule packs)94.46%52.97%67.87%83.88%
NVIDIA SkillSpector48.11%62.62%54.42%28.72%

Any finding at HIGH or above

ToolRecallPrecisionF1False-positive rate
Cisco skill-scanner (core rules)31.23%95.38%47.06%1.51%
Cisco skill-scanner (all rule packs)91.44%56.45%69.81%70.53%
NVIDIA SkillSpector30.48%75.62%43.45%9.82%

Each tool's own shipped install gate

ToolRecallPrecisionF1False-positive rate
Cisco skill-scanner (core rules)31.23%95.38%47.06%1.51%
Cisco skill-scanner (all rule packs)91.44%56.45%69.81%70.53%
NVIDIA SkillSpector8.56%91.89%15.67%0.76%

Block on any finding at all

ToolRecallPrecisionF1False-positive rate
Cisco skill-scanner (core rules)81.86%47.86%60.41%89.17%
Cisco skill-scanner (all rule packs)99.75%50.06%66.67%99.50%
NVIDIA SkillSpector50.13%63.17%55.90%29.22%

Cost of a scan

ToolMean seconds per skillInput tokensOutput tokens
Cisco skill-scanner (core rules)0.017not instrumentednot instrumented
Cisco skill-scanner (all rule packs)0.280not instrumentednot instrumented
NVIDIA SkillSpector2.399not instrumentednot instrumented

Integrity

ToolRowsUsableErrorsCapability degradedReported partial
Cisco skill-scanner (core rules)8008000018
Cisco skill-scanner (all rule packs)8008000018
NVIDIA SkillSpector80079406674

OpenSkillRisk

335 records scored on both tools · 182 contextually_risky, 153 obviously_malicious

Any finding at MEDIUM or above

ToolRecallPrecisionF1False-positive rate
Cisco skill-scanner (core rules)25.07%no harmless classno harmless classno harmless class
Cisco skill-scanner (all rule packs)96.42%no harmless classno harmless classno harmless class
NVIDIA SkillSpector77.91%no harmless classno harmless classno harmless class

Any finding at HIGH or above

ToolRecallPrecisionF1False-positive rate
Cisco skill-scanner (core rules)21.49%no harmless classno harmless classno harmless class
Cisco skill-scanner (all rule packs)91.64%no harmless classno harmless classno harmless class
NVIDIA SkillSpector62.69%no harmless classno harmless classno harmless class

Each tool's own shipped install gate

ToolRecallPrecisionF1False-positive rate
Cisco skill-scanner (core rules)21.49%no harmless classno harmless classno harmless class
Cisco skill-scanner (all rule packs)91.64%no harmless classno harmless classno harmless class
NVIDIA SkillSpector33.43%no harmless classno harmless classno harmless class

Block on any finding at all

ToolRecallPrecisionF1False-positive rate
Cisco skill-scanner (core rules)97.61%no harmless classno harmless classno harmless class
Cisco skill-scanner (all rule packs)100.00%no harmless classno harmless classno harmless class
NVIDIA SkillSpector78.21%no harmless classno harmless classno harmless class

Cost of a scan

ToolMean seconds per skillInput tokensOutput tokens
Cisco skill-scanner (core rules)0.018not instrumentednot instrumented
Cisco skill-scanner (all rule packs)0.266not instrumentednot instrumented
NVIDIA SkillSpector2.457not instrumentednot instrumented

Integrity

ToolRowsUsableErrorsCapability degradedReported partial
Cisco skill-scanner (core rules)374374005
Cisco skill-scanner (all rule packs)3743650914
NVIDIA SkillSpector374343031302

HarmfulSkillBench

197 records scored on both tools · 197 unlabelled

Any finding at MEDIUM or above

ToolRecallPrecisionF1False-positive rate
Cisco skill-scanner (core rules)4.06%not permittednot permittednot permitted
Cisco skill-scanner (all rule packs)90.86%not permittednot permittednot permitted
NVIDIA SkillSpector56.85%not permittednot permittednot permitted

Any finding at HIGH or above

ToolRecallPrecisionF1False-positive rate
Cisco skill-scanner (core rules)3.55%not permittednot permittednot permitted
Cisco skill-scanner (all rule packs)83.76%not permittednot permittednot permitted
NVIDIA SkillSpector17.26%not permittednot permittednot permitted

Each tool's own shipped install gate

ToolRecallPrecisionF1False-positive rate
Cisco skill-scanner (core rules)3.55%not permittednot permittednot permitted
Cisco skill-scanner (all rule packs)83.76%not permittednot permittednot permitted
NVIDIA SkillSpector2.54%not permittednot permittednot permitted

Block on any finding at all

ToolRecallPrecisionF1False-positive rate
Cisco skill-scanner (core rules)96.45%not permittednot permittednot permitted
Cisco skill-scanner (all rule packs)100.00%not permittednot permittednot permitted
NVIDIA SkillSpector56.85%not permittednot permittednot permitted

Cost of a scan

ToolMean seconds per skillInput tokensOutput tokens
Cisco skill-scanner (core rules)0.017not instrumentednot instrumented
Cisco skill-scanner (all rule packs)0.325not instrumentednot instrumented
NVIDIA SkillSpector2.400not instrumentednot instrumented

Integrity

ToolRowsUsableErrorsCapability degradedReported partial
Cisco skill-scanner (core rules)2002000019
Cisco skill-scanner (all rule packs)2002000019
NVIDIA SkillSpector20019703160

Limits on what these numbers support

Our rule packs were tuned against MaliciousSkillBench during development. SkillSpector has not seen it. The source-disjoint split and the corpora outside MSB are therefore the ones that carry weight, and figures on the balanced subset should be read as home ground. SkillSpector was tuned on a private 31,000-skill set that is not published, so contamination can be bounded in one direction only and is disclosed rather than measured in the other.

Enabling every rule pack this scanner ships is not its strongest configuration. On an 80/80 sample of the source-disjoint split, all packs raise recall from 8.8% to 73.8% and raise the benign flag rate from 7.5% to 92.5%. That is a triage setting rather than a gating one, so the shipped core profile is what appears above; the ATR pack is also excluded from source-disjoint claims by the corpus terms.

One SkillSpector capability is absent from these runs. Its --transitive mode follows references out of a skill, which on this corpus would mean fetching attacker-controlled URLs, and the corpus safety defaults forbid it. Its supply-chain recall would likely be higher with that enabled, so the figures here understate it on that axis.

MaliciousSkillBench stores each record as one SKILL.md. Analyzers that operate on bundled scripts or MCP manifests have nothing to read, on both sides, and report not_applicable rather than failing. Three of SkillSpector's 27 analyzers are inert on that corpus for this reason. OpenSkillRisk and HarmfulSkillBench carry the multi-file signal.

SkillSpector reports analysis as partial on most MSB records. Coverage is 100%, nothing is left uninspected, and the ledger exceptions are non-fatal: each SKILL.md references files the corpus does not ship, and the tool says so. That is a reporting capability this scanner does not have, and the partial rate is recorded in the integrity table rather than treated as a defect or used to exclude rows.

Finding counts are not compared directly. SkillSpector applies diminishing returns per rule and caps at three occurrences, so its issue count and this scanner's finding count measure different things. Distinct rules fired and records flagged are used instead.

Run notes