macleodlabs-ai

Final completion/sanity check — audit the current code with a team of subagents to prove it is really finished (no stubs/TODOs/no-ops), plus correctness bugs and leanness issues, then fix them and verify. Use when the user types /go, or asks to sanity-check completion, audit-and-fix, review-and-fix, clean up, or harden the current branch/diff.

Draft replies to incoming messages (email, Slack, tickets, forms) in the user's own voice, and check inbound or outbound work against a quality contract before it is sent. Use whenever the user asks to reply to, respond to, draft, or triage a message on a comms channel, or to review whether a piece of work is ready to send. In learning mode, improves at matching the user's voice over time from how they edit the drafts it produces.

Set up or repair an isolated per-client Claude Code config directory with a two-tier CLAUDE.md memory system (a shared global file imported into a client-local file), a client-aware statusline that labels which account or billing entity the session belongs to, and optional shell launcher functions (named cc-clientname) that set the config-dir environment variable automatically plus neuter the bare claude command. Use this skill whenever the user wants to configure a new client or work config dir, add shared-plus-local memory across multiple Claude Code accounts, stop sessions from looking identical in the footer, set up per-client launch commands so they don't have to export the config-dir variable by hand, fix a config dir that's missing the shared import or statusline, or asks to "set up this Claude", "bootstrap my config", or "run the client setup". Safe and idempotent — re-running only changes what's missing or has drifted, so reach for it even when unsure whether a dir is already configured.