Skill Discovery and Use over A2A
Guide Demo + Screencast

Establish context. Discover endpoint skills. Select the skill that matters.

The demo starts with the principal establishing operating identity, then adds preferences as behavioral context, then invokes the A2A client skill to discover and select capabilities from the chosen endpoint.

Demo guide and screencast assembled by a2a-client, screencast-recorder, and GPT-5 Chat Codex on behalf of Kingsley Idehen.

I am the principal. Codex is the agent. The A2A skill discovers what my chosen endpoint can do, then helps select the skill of interest.
Voiceover Screencast

Playable Demo Recording

The final cut should follow the simple story: principal identity, agent relationship, preferences, A2A invocation, endpoint skill discovery, and selected skill. The MP4 includes embedded English captions; the page also supplies the VTT track for browser caption controls.

Scene 1

Principal-Agent Context

The interaction opens with whoami?. That establishes the operating identity for the session and makes the relationship explicit: Kingsley is the principal, and Codex is the agent acting in the session on his behalf.

user> whoami? User (principal) • Name: Kingsley Uyi Idehen • Email: kidehen@openlinksw.com • Role: Founder & CEO, OpenLink Software • WebID: link-in-bio credentials NetID verified • Delegation: local profiles corroborate agent acting on behalf of user Agent (me) • Model: GPT-5 Chat Codex • Environment: Codex Desktop • Memory store: agent-rdf-memory/ • Behavioral contract: preferences.ttl • Agent WebID: link-in-bio agent NetID verified
Scene 2

Preferences Add Behavioral Context

The second opening interaction is What are my preferences?. Those preferences add session context that guides agent behavior: memory use, endpoint-aware workflows, Linked Data handling, and compact hyperlink-first presentation.

#CategoryStepsPrimary HowTo Files
1Identity, WebID & Delegation14agent-identity.ttl, verified-identity.ttl
2Memory Management & Session Governance20session-governance.ttl, memory-protocol-gate.ttl
3Artifact Output Routing11artifact-routing.ttl, model-identity-resolution.ttl
4RDF Authoring & Entity Resolution21canonical-entity-iri-denotation.ttl, entity-link-placement.ttl
5HTML Infographic & KG Explorer49infographic-authoring.ttl, kg-explorer-d3-patterns.ttl
6Skill Chain & Workflows7skill-invocation.ttl, uriburner-oauth-authcode-flow.ttl
7Resolvable IRI Payload Presentation1resolvable-iri-payload-presentation.ttl

Session-relevant preference: when a payload includes a label and a resolvable IRI, present the label as the hyperlink and omit the redundant raw IRI column by default.

Scene 3

Invoke the A2A Skill

Once identity and preferences are established, the user invokes the A2A client skill. Its job is to discover endpoint capabilities and help choose the skill of interest.

IdentityWho is the principal?
DelegationWho is the agent?
PreferencesWhat behavior should guide the session?
DiscoverWhat skills does the chosen endpoint offer?
SelectWhich skill is of interest?
Scenes 4-5

Endpoint Discovery

The A2A skill discovers capabilities from endpoint Agent Cards. Endpoint choices remain user-controlled; the client does not hardcode a CNAME as the default.

EndpointAgent CardA2A URLAuthStreamingSkillsNotes
https://netid-qa.openlinksw.com404N/AN/AN/AN/ANo card at /.well-known/agent.json
https://netid-qa.openlinksw.com:8443OKhttps://netid-qa.openlinksw.com:8443/chat/api/a2aOAuth2true9Working NetID QA base
https://linkeddata.uriburner.comOKhttps://linkeddata.uriburner.com/chat/api/a2aOAuth2true13URIBurner endpoint
https://demo.openlinksw.comOKhttps://demo.openlinksw.com/chat/api/a2aOAuth2true37Demo endpoint
Scenes 6-7

Skills by Endpoint

The crux of the demo is discovery of skills and selection of a skill of interest on the chosen endpoint. Data Twingler on URIBurner is highlighted because that is the selected skill for the session.

EndpointSkill IDSkill Name
NetID QA :8443system-data-twingler-configOpenLink Data Twingler v2.0.4
URIBurnersystem-data-twingler-configOpenLink Data Twingler v2.0.4
URIBurnersystem-virtuoso-support-assistant-configVirtuoso Support Agent v1.1.71
URIBurnersystem-opml-rss-news-readerOPML and RSS News Reader Assistant v1.0.40
URIBurnersystem-md~linked-data-agent-2linked-data-agent-2
Demosystem-data-twingler-configOpenLink Data Twingler v2.0.64
Demosystem-virtuoso-support-agent-canonicalVirtuoso Support Agent v1.2.09
Demosystem-fint-assistant-sparqlFint Demo Agent (SPARQL) v1.0.1
Full discovery table is available in the session transcript

The guide version should render the complete skills-by-endpoint table as an expandable section. The screencast can pause on the highlighted URIBurner Data Twingler row.

Optional Continuation

After Selection: Interaction Evidence

After skill selection, the workflow can continue into authorization and task execution. These details are useful QA evidence, but they support the story rather than define it.

task OpenLink Data Twingler v2.0.4: Using DBpedia, list movies by Spike Lee. unauthenticated send HTTP 401 oauth browser login completed token .opal-a2a.env written retry authenticated request reached OPAL backend Data Twingler missing-parameter error

What Worked

Discovery OAuth Token capture Authenticated reach

What Needs Fixing

The Data Twingler backend route returned a missing-parameter error after authentication. The screencast should present this as a QA finding, not as a failure of the discovery/auth workflow.

Supporting Output

Result Presentation Preference

The DBpedia result remains useful as supporting evidence for the presentation preference: when a payload includes a title and a resolvable IRI, the title should be the hyperlink and the redundant IRI column should be omitted.

Close

Takeaway

ResultStatusMeaning
Principal identity establishedDoneThe session knows who the agent is working for.
Agent relationship establishedDoneThe agent role is explicit in the session context.
Preferences loadedDoneBehavioral context guides the work.
A2A skill invokedDoneThe skill discovers capabilities from endpoint Agent Cards.
Skill of interest selectedDoneURIBurner Data Twingler v2.0.4 is the selected skill.
Task execution evidenceSupporting QAOAuth worked; backend route exposed a parameter issue for follow-up.