← ClaudeAtlas

grok-promptinglisted

Internal guidance for composing Grok (xAI) prompts for coding, review, diagnosis, and research tasks inside the polycli plugin
bbingz/polycli · ★ 0 · AI & Automation · score 66
Install: claude install-skill bbingz/polycli
# grok-prompting Guidance for prompts sent to the Grok Build CLI via `/polycli:ask|rescue|review --provider grok`. ## When to reach for grok - A fast, capable second opinion on coding/review tasks; `grok-composer-2.5-fast` (Compose 2.5) is the default and is tuned for code. Use `-m grok-build` for the heavier build-agent model. - Reasoning-heavy diagnosis: pass `--effort high` (or `xhigh`/`max`) so grok spends more reasoning. ## Prompt shape - grok one-shot mode (`-p`) returns a single visible answer — write self-contained prompts; it does not carry conversation unless you `--resume <id>` / `--continue`. - For `/review`, polycli already forces `--permission-mode plan` (read-only) and the review prompt forbids tools/edits — keep the focus terse (the diff is supplied for you). - grok emits a `thought` (reasoning) channel separate from the answer `text`; ask for a concise final answer so the visible `text` is self-sufficient, not only reasoning. ## Avoid - Don't ask grok to run long multi-step tool loops in `ask` — `-p` is a one-shot print mode; for agentic work prefer a provider with a persistent session, or expect a single pass. - Don't paste secrets; polycli redacts argv in the run ledger but the prompt body is sent verbatim.