Agent skills

plous-debug

Use plous-debug in Codex or Claude Code to investigate Kubernetes failures, test a suspected cause, and check a deployed fix.

Your agent keeps its usual repository tools. Plous adds cluster tools through MCP. The Plous terminal UI does not need to be running.

Install

Install Plous, then run the command for your agent:

sh

plous setup codex

For Claude Code:

sh

plous setup claude

Setup asks for your kubeconfig and context. It checks the API connection, installs the skill, and configures MCP. It finds the Plous executable automatically.

Restart or reconnect your agent. Invoke   $plous-debug   in Codex or   /plous-debug   in Claude Code.

To supply the target without prompts:

sh

plous setup codex \ --kubeconfig /path/to/kubeconfig \ --context staging \ --namespace shop

Use   claude   in place of   codex   for Claude Code. Both use the client's existing authentication.

Files installed

Client

Skill

MCP configuration

Codex

  ~/.agents/skills/plous-debug/SKILL.md

  ~/.codex/config.toml  , or   $CODEX_HOME/config.toml

Claude Code

  ~/.claude/skills/plous-debug/SKILL.md

  ~/.claude.json

Setup preserves other MCP entries. It stops if an unrelated skill or unmanaged   plous_debug   entry already uses the destination. Rerun setup to update an installation managed by Plous.

  --home PATH   chooses a separate installation home.   --skip-check   skips the API connection test but still validates the local target settings.

Investigate, reproduce, and verify

Name the symptom, context, and namespace:

Use plous-debug. Checkout returns 502 in staging/shop. Check its Pods, events, and logs. Find the cause and test a fix in this repository.

The agent can read Kubernetes objects, events, logs, metrics, and available Helm, Argo CD, and Flux data. See the tool list.

Deploy through your usual process. Then ask:

Verify the checkout fix in staging/shop. Check the running image and new errors. Repeat the failing request if diagnostics are enabled.

A passing local test does not confirm that a fix is deployed.

Enable active debugging

Reads are enabled by default. Add   --allow-exec   during setup to allow commands in existing containers.

sh

plous setup codex --allow-exec

Add   --diagnostic-image IMAGE   to allow temporary HTTP, DNS, and TCP tests. Both options apply to the selected default namespace and require Kubernetes permissions.

See Active diagnostics for examples, image requirements, and cleanup.

Scope, evidence, and lifecycle

The cluster, default namespace, and diagnostic permissions stay fixed until the agent reconnects. Reads can use other namespaces permitted by your Kubernetes credentials.

Each MCP process saves its own tool evidence. It does not search earlier standalone sessions automatically. Logs and command output may be sent to the agent's model provider. Kubeconfig credentials stay with Plous.

See Scope and saved evidence for storage locations and cancellation behavior.

Remove the integration

Delete the installed   plous-debug   skill directory and remove its MCP entry:

  • Codex: remove the block between   BEGIN PLOUS DEBUG MCP   and   END PLOUS DEBUG MCP  .

  • Claude Code: remove   mcpServers.plous_debug  .

Reconnect the client. Saved evidence remains until you delete the standalone profiles.

Next steps

Try Investigate a restarting Pod. Use MCP setup for a connection without the skill. The skill's source is skills/plous-debug.

View source

Plous · Open source · Early development

Search Plous docs

Search the documentation.

Documentation versions