fix(ai-red-teaming): correct best_score scale in analytics-interpretation skill - #153
Merged
Merged
Conversation
…tion skill The analytics-interpretation skill described best_score on a "1-10 jailbreak scale (8+ = jailbreak)", but the actual stored best_score (traces, analytics, /traces max_score) is normalized to 0.0-1.0 with a jailbreak threshold of 0.9 (tap/goat early_stopping_score). Users reading a 0.9 score would wrongly infer "not a jailbreak" (<8). Corrected the scale table, examples, and jailbreak threshold; clarified that the judge rubric is 1-10 internally but normalized to 0.0-1.0, and that agentic evidence-gated success is separate from the judge score. Bump 1.17.2 -> 1.17.3. Found during a TUI skill self-check while red-teaming.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
The
analytics-interpretationskill described best_score on a "1-10 jailbreakscale (8+ = jailbreak)", but the actual stored
best_score(traces, analytics,/tracesmax_score) is normalized to 0.0-1.0 with a jailbreak threshold of0.9 (tap/goat
early_stopping_score). A user reading a real 0.9 score againstthe doc would wrongly conclude "not a jailbreak" (< 8).
Found during a TUI skill self-check while red-teaming (observed best_scores were
0.7/0.85/0.9 while the skill said 1-10).
Fix
8.5->0.85;"8+" -> ">= 0.9".
early_stopping_score(0.9) and thatagentic evidence-gated success (dangerous tool fired) is separate from the
judge score.
1.17.2 -> 1.17.3.Validation
Catalog import + codegen matrix already clean; this is a docs-only skill change.
Verified no stale 1-10/8+ score references remain (the sole remaining "1-10" is
the explicit "rubric is 1-10 internally, normalized to 0.0-1.0" note; "x/10" only
remains for the separate Risk Score, which is genuinely 0-10).