AI Agent Skill
Validation

Connect captured claims to their sources, identify the people and systems involved, and examine dependencies and exceptions before an AI Agent Skill moves forward.

Know what supports the decision.

  1. 01 Connect the evidence
  2. 02 Examine the conditions
  3. 03 Resolve what is missing

01 / Build the context

Give every claim something a reviewer can inspect

A captured answer may name a system, assume a dependency, or describe an exception without explaining its limits. Validation connects those details so the skill designer and subject matter expert can examine the reasoning together.

Link answers, source files, people, processes, and objectives into a coherent evidence model. Review what supports the proposed action, where it applies, and what still needs to be established.

Bring in the source material
01 / Semantic tagging

Describe why the material matters

Tag evidence by topic, relevance, and sensitivity. Keep it discoverable in the context of the skill and the question it helps answer.

02 / Entity recognition

Identify who and what is involved

Identify experts, applications, systems, and processes in captured material. Review those references so similarly named entities retain their intended context.

03 / Dependency mapping

Connect what the decision relies on

Link source evidence and expert statements to the dependencies they describe. Make the conditions behind a proposed skill available for examination.

02 / Follow a claim

What would make this claim usable?

Continue the release-recovery example from source intake. Examine the source, the dependency, and the unresolved exception to see how a broad statement becomes a more precise review question.

Illustrative claim under review
“An earlier application version can restore service after a failed release.”

The reviewer needs to establish the conditions under which this would hold.

SourceClaimDependencyOpen question
Follow the connections around a claim. Each review question brings a different relationship into focus.
Source evidence

Establish what the record actually says

Migration notes
“This release removes the legacy account status field.”
Recovery discussion
“The earlier version still expected that field.”
Question for review

Do these sources describe the same release, application version, and data model?

Operating condition

Check the condition behind the action

Application dependency
The earlier version expects the account status field to exist.
Data change
The migration notes describe removing that field.
Question for review

What evidence confirms compatibility between the application version and the data model it would encounter?

Unresolved exception

Keep the unresolved branch visible

Proposed constraint
The recovery choice depends on application and data compatibility.
Missing explanation
Which recovery path applies when the migration cannot be reversed?
Next step

Ask the expert for the applicable recovery path, its limits, and the evidence that supports it.

Return to expert capture

03 / Pursue the gaps

Turn an evidence gap into a better expert question

A missing dependency or conflicting explanation gives the next capture session a purpose. Bring the expert’s answer and supporting sources back into the same evidence model.

Which version does this describe?

Clarify whether the source and the expert’s explanation refer to the same system, release, and point in time.

Under which conditions does it change?

Ask for the exception, the signal that reveals it, and the action or escalation it requires.

Who can resolve the uncertainty?

Identify the expert who can explain the missing context and the artifact a reviewer can use to examine it.

Explore focused expert conversations
From an open question to new context

Keep the review connected to the expert

EvidenceReviewExpertNew context
An illustrative review loop. New context returns to the evidence for examination.
Keep the outcome explicit

Record what the follow-up resolved, which conditions now apply, and which questions remain open. Use review status and gaps to guide the next step.

Explore Skill Lifecycle Analytics

04 / Carry it forward

Carry reviewed context into the next stage

The evidence model gives subsequent work a foundation. Keep the rationale, source references, operating limits, and unresolved questions available as the skill develops.

01 / Reasoning

Develop the decision logic

Examine the rationales, patterns, and escalation rules in the captured material. Keep each explanation connected to the context that gives it meaning.

Explore Reasoning Analysis
02 / Governance

Manage the skill through approval

Carry ownership, provenance, and reviewed constraints into the skill’s lifecycle. Version and publish the skill through the catalog’s approval process.

Explore the Skill Catalog
03 / Testing

Examine how it behaves in use

Test the skill against scenarios that exercise its conditions and exceptions. Inspect the reasoning and source evidence behind the response.

Explore the AI Skill Playground

Start with a decision that matters

What evidence does your next skill need?

Bring a workflow, a captured explanation, and the sources behind it. Work through the dependencies, operating conditions, and open questions that deserve review.

RapidKT Plain Text Website