/bias
Investigate bias across AI training data, training processes, dependencies and product behavior.
Diego · AI chatbotAI evaluationsWhen to use it
Uneven representation or outcomes could be hidden by the aggregate.
Before you start
Define relevant populations and justified comparisons. Inspect data and product behavior, report uncertainty and slice-specific failures, and distinguish disparities from conclusions about intent or legal compliance.
Try this in your coding agent.
/bias Review the recommendation evaluation for uneven outcomes across supported languages.Replace sample paths with your project. Slash names are shorthand; your host may use a plugin prefix or a skill picker. Invocation help →
ILLUSTRATIVE WALKTHROUGH
Recommendation-evaluation balance
An expanded teaching example, not a recorded or live execution. The report stays visible while the replay explains how to read it.
- 01Read context
- 02Analyze
- 03Show evidence
- 04Report
Coding-agent conversation
I’m reviewing recommendation-evaluation balance. I’ll keep the scope and permissions explicit.
Analyze: Review the recommendation evaluation for uneven outcomes across supported languages.
Here is the evidence trail to inspect: Compare supported-language coverage and error types using synthetic labeled slices.
Language slices are imbalanced, so the aggregate score is misleading. A disparity signal is not proof of discriminatory intent or legal noncompliance.
Example report
Illustration
TestBucks course screen with illustrative analysis. The screenshot supplies context; the demo does not establish a newly observed bug.
- Question / intent
- Reference evidence
- Scoring & uncertainty
Illustrated outcome
Language slices are imbalanced, so the aggregate score is misleading.
Evidence to inspect
Compare supported-language coverage and error types using synthetic labeled slices.
Still untested / not established
A disparity signal is not proof of discriminatory intent or legal noncompliance.
Ready. Read the full example or play its steps.
What a useful result contains
A bounded claim
For recommendation-evaluation balance, state the target, revision, environment and actual work performed. Keep planned, executed, blocked and untested work separate.
An inspectable trail
Compare supported-language coverage and error types using synthetic labeled slices. In a real run, retain the source, trace, screenshot or fixture behind the conclusion.
The qualification
A disparity signal is not proof of discriminatory intent or legal noncompliance.
The next decision
Decide whether the evidence calls for a focused check, a reviewed fix or more context. Do not convert a plausible example into a verified finding.
Adapt the example to your project
- Replace recommendation-evaluation balance with a real feature, path or decision in your project.
- State the expected rule and its source. Include safe fixtures, constraints and a budget where appropriate.
- Ask for exact evidence and remaining gaps. Approve consequential actions separately.
/jay Review the /bias result. Separate observed evidence, assumptions and untested scope, then recommend the next useful check.Where to go next
/carbon-eval →/carbon-statistical-review →Content checked against the local command definition: skills/bias/SKILL.md. Examples adapted from the homepage’s TestBucks walkthroughs. Availability is edition- and host-dependent.