example: Deploy Claude Code as a Kagenti-managed agent

[jenna-winkler]

Description

Deploy a Claude Code container on Kagenti and validate that the platform's core capabilities (identity and governed tool access) work with a harness-style agent.

Today all Kagenti examples are framework-based agents (e.g., LangGraph) that follow a build-deploy-discover lifecycle. Claude Code is different - there's no build step, no A2A agent card, no framework code. You configure it with a system prompt, skills, and tools. This example validates whether Kagenti's platform value applies to that model.

What the example demonstrates

• Credentials - Anthropic API key stored as a K8s Secret, injected via env var.
• Identity (SPIFFE/SPIRE) - Claude Code pod gets a SPIFFE workload identity like any other Kagenti agent.
• Tool access (MCP Gateway) - Claude Code connects to platform-managed MCP tools through the MCP Gateway.
• User interaction - Users talk to the agent via Slack/Discord bot.
• Session persistence - Agent home directory backed by a PVC so Claude Code's native session storage survives pod restarts.

Out of scope (follow-up)

• Observability (Phoenix tracing for harness sessions)
• UI integration (agent visible and testable from Kagenti dashboard)
• Scheduling and triggers (cron, webhook, GitHub Actions)

[jimcadden]

@esnible this seems aligned with what we spoke about today. Can you pull in Rong and Jarek too?

[github-actions]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed in 14 days if no further activity occurs. If this issue is still relevant, please comment or remove the stale label.

[github-actions]

This issue was closed because it has been stale for 14 days with no activity. Feel free to reopen if this is still relevant.

[esnible]

Delivered through #531