Nothing installed here (no populated .claude/skills / .claude/agents, and no global install to target) -> /claude-stack:setup (fresh install from scratch).
The stack is installed and the ask is a plain refresh (no items named) -> /claude-stack:update (refresh everything installed + prune what upstream removed).
The stack is installed and the user wants to adjust it (add or drop items, change the selection) -> /claude-stack:configure.
The stack is installed and the user wants it reconciled TO THIS PROJECT - prune what the project's frameworks do not use (WPF artifacts in a web-only repo, the data vertical with no SQL) AND add the detected stacks' artifacts that are missing -> /claude-stack:validate (project mode only).
The ask is to SEE what is installed - inventory, versions, capture dates, 'what do I have?' - with nothing to change -> /claude-stack:status (read-only tables, per area or all).
The install just finished and the ask is 'what now?' (session reload, gitignore lines, serena setup, which captures in which order) -> not a command: walk them through ${CLAUDE_PLUGIN_ROOT}/references/post-install.md.

Answer with the command plus one line naming the state you found (for example: 'no .claude/skills here - run /claude-stack:setup'). When more than one reading is plausible, put the candidates through AskUserQuestion - one option per plausible command, the likeliest marked Recommended - instead of guessing or asking in prose.

claude-stack

/claude-stack - the router