CARBON — Engineering ConfidenceUser Guide
CHAPTER 41 / 60 Pro

/carbon-incident

Respond to an AI production incident by containing impact, preserving prompts and traces, identifying contributing conditions, assessing customer harm, and promoting the failure into durable eval cases.

Jay · Your AI test managerProof & release confidence

When to use it

An AI feature has produced a real harmful or incorrect outcome.

Before you start

Preserve traces and bound known impact before changing systems. Propose containment for approval, reproduce safely, and turn the failure into a regression case. Do not infer permission to contact users or alter production.

Try this in your coding agent.

/carbon-incident Investigate incorrect chatbot answers and propose containment steps for approval.

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

Incorrect support answers

An expanded teaching example, not a recorded or live execution. The report stays visible while the replay explains how to read it.

  1. 01Preserve
  2. 02Assess impact
  3. 03Reproduce
  4. 04Propose containment

Coding-agent conversation

Preserve

I’m reviewing incorrect support answers. I’ll keep the scope and permissions explicit.

Assess impact

Separate known affected examples from possible impact; preserve the evidence.

Reproduce

Here is the evidence trail to inspect: Preserve examples → bound impact → propose disabling affected answer path → request approval.

Propose containment

Jay proposes containment and a focused reproduction plan. No production mitigation or external message is executed.

CARBON

Example report

Illustration
Failure → evidence → recovery
Failure → evidence → recovery — TestBucks teaching screen

TestBucks course screen with illustrative analysis. The screenshot supplies context; the demo does not establish a newly observed bug.

  1. Failed request
  2. Preserved state
  3. Safe retry

Illustrated outcome

Jay proposes containment and a focused reproduction plan.

Evidence to inspect

Preserve examples → bound impact → propose disabling affected answer path → request approval.

Still untested / not established

No production mitigation or external message is executed.

Ready. Read the full example or play its steps.

What a useful result contains

A bounded claim

For incorrect support answers, state the target, revision, environment and actual work performed. Keep planned, executed, blocked and untested work separate.

An inspectable trail

Preserve examples → bound impact → propose disabling affected answer path → request approval. In a real run, retain the source, trace, screenshot or fixture behind the conclusion.

The qualification

No production mitigation or external message is executed.

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

  1. Replace incorrect support answers with a real feature, path or decision in your project.
  2. State the expected rule and its source. Include safe fixtures, constraints and a budget where appropriate.
  3. Ask for exact evidence and remaining gaps. Approve consequential actions separately.
/jay Review the /carbon-incident result. Separate observed evidence, assumptions and untested scope, then recommend the next useful check.

Where to go next

/carbon-eval →/carbon-generate →

Content checked against the local command definition: skills/carbon-incident/SKILL.md. Examples adapted from the homepage’s TestBucks walkthroughs. Availability is edition- and host-dependent.