/forget

The /forget skill wipes kstack’s local state on your machine. Over time kstack builds up a working memory of your clusters — recent query results, detected integrations, resource fingerprints, and baselines it uses to detect anomolies. This skill is a forced clearing of that data so the agent can start with a clean slate.

This skill does not touch anything in the cluster. For removing kstack-managed resources from the cluster itself, see /cleanup.

/forget                      # forget everything for the current context
/forget --context prod       # forget everything for the 'prod' context
/forget --all                # forget everything for all contexts

User-invoked only

/forget ships with disable-model-invocation: true. The agent will never wipe local state on its own — it only runs when you explicitly type /forget, so you don’t lose cached context unexpectedly.

---

What it clears

State lives under ~/.config/kstack/, partitioned per kubeconfig context — forgetting staging never affects prod.

Local state

• Cache (~/.config/kstack/cache/<context>/) — recent query results, log buffers, dedup tables, in-flight watcher state. Cheap to rebuild; cleared freely.
• Learned state (~/.config/kstack/state/<context>/) — detected integrations, resource fingerprints, baselines, per-cluster preferences. Rebuilt on next use, but may take a few interactions to fully re-form.

By default the skill clears both for the current context.

---

When to run it

• The cluster was rebuilt or migrated and its old identity no longer matches what kstack remembers.
• You want kstack to re-learn from scratch — baselines feel stale, or an earlier session taught it something wrong.
• You’re handing the machine off and want no cluster-specific state left behind.

---

Options

--all

Clear cache and learned state for every context, not just the current one

Global flags from Overview also apply.