BF benchmark-robustness-auditor
Offline, defensive robustness auditor for LLM benchmarks: n-gram exact + shingle-Jaccard paraphrase contamination, temporal pre/post-cutoff gaps, TS-Guessing above-chance detection, option-letter selection bias (chi2), few-shot curve noise, LLM-judge position/verbosity/rubric-echo bias and hidden-instruction payload detection, paired McNemar + Wilson + deterministic bootstrap for score comparisons, WORKED mitigations (permutation majority ensemble, blind content normalization), documented 0-100 severity formula, hash-chained per-target history ledger with trend deltas. Findings cite a STATIC 17-id exploit catalogue with explicit computable flags — invisible exploit classes are declared, never fabricated. 100% stdlib python3. NO network, NO telemetry. Defense/auditing only.
Offline, defensive robustness auditor for LLM benchmarks: n-gram exact + shingle-Jaccard paraphrase contamination, temporal pre/post-cutoff gaps, TS-Guessing…
As a process F 44/100 · Will not run — References files that are not bundled: scripts/benchscan.py
How to improve
- Say in the description WHEN to use the skill ("use when…", example requests): that is the agent's main cue.
- The text references files that are not there: add them or drop the references.
- Your own cases (evals/evals.json, 4–6 real requests with expected answers): the full check would then run those instead of a model-drafted suite.
- A spec.yaml with trigger phrases and assertions — a behaviour contract for CI; `skilltest init` writes a template.
Guard findings · 0
✓ No critical or high findings
Files scanned: 0. Evidence is masked. Grey chips explain why severity was lowered.
Against the Agent Skills spec
- warning
description-no-whendescription does not say WHEN to use the skill (no "use when") - warning
missing-refreference to a missing file: scripts/benchscan.py - note
frontmatter-keyunknown frontmatter key "topics"
Process rating: all ten parameters 44/100
- 0Tools and files. 1 referenced file(s) missing: scripts/benchscan.py
- 0Result and completion. Does not say what the result is
- 0Failures and branches. Linear process with no failure handling
- 0Progress reporting. Says nothing while it works
- 30Running it twice. 1 mutating operations with no state check
- 50When it triggers. No condition that starts the skill
- 70Inputs and preconditions. Inputs and preconditions are listed
- 100Steps. 12 steps
- 100Consistency. Name and required fields are in place
- 100Execution cost. Instruction body is 895 tokens
Everything here is measured from the skill text rather than judged by a model, so the numbers are checkable. A parameter weighs more when it is a more common reason for the process to stall.
Quality signals
- +5Description has no quoted example phrases that should trigger the skill
- +4Description does not say when NOT to use the skill (false activations)
- +3Output format is not stated: the model decides each time
- +4No input/output examples
- +1No license
- +2Single-language instructions
- +3Description length 783: enough signal without eating the budget
- +4Structure: 4 headings
- +3Step-by-step instructions: 12 items
Quality base 70; lint remarks subtract, signals add up to 100. Result: 61.