The Krevaya Claude Skill for disciplined content ops

A Claude Code skill that runs a review-first session over Krevaya's MCP tools, from results to drafts to scheduling

Written by the Krevaya team · Updated

The Krevaya Claude Skill teaches Claude Code a disciplined way to run your account through Krevaya's MCP tools: read your results first, triage drafts with you, then schedule. It assumes Claude Code is already connected to Krevaya's MCP server.

What a session looks like

  1. Read before acting. Claude opens with get_queue_summary, get_audit (your latest weekly audit), and get_analytics for the last 30 days, then sums up in three lines: what's waiting, what the audit recommends, and what performed. Before it judges any copy, it reads get_brand_memory.
  2. Triage drafts with you. It lists drafts with list_posts and gives a recommendation for each, grounded in your numbers. It never calls approve_post or reject_post until you have seen the draft and said yes, one by one or explicitly in bulk. Every rejection carries a reason and a code: off_voice, wrong_topic, too_salesy, factually_off, bad_hook, too_long, or other. It stars a draft with favorite_post only when you say it's exactly right.
  3. Schedule. Approved posts go out with schedule_post and no time, which takes the channel's next free posting slot. It uses get_best_times only when you want a specific day.
  4. Generate only into a thin queue. If drafts are scarce and you ask for more, it uses draft_week (and polls job_status) or repurpose for a link or text, picks topics from list_trends and the audit, and explains why.

It ends with a short report: how many drafts it approved, rejected, and scheduled, what it generated, and the one thing to watch next.

How it treats your brand memory

  • update_brand_memory records a change as something you stated and locks it, so Claude only writes what you actually told it.
  • It sends the version it read. If Krevaya or someone else changed the memory in between, it re-reads and reapplies instead of overwriting.
  • On a new account, import_history pulls your own published posts from Instagram, Facebook, Threads, and Bluesky so Krevaya learns your voice. It makes no AI call, and a second run imports nothing new.

Permissions it needs

The AI assistant key preset covers everything the skill does. Reading needs View content; approving, rejecting, starring, and scheduling need Create & publish posts; draft_week, repurpose, and brand memory edits need Generate content. draft_week and repurpose use AI credits. A Read only key turns the skill into an analyst that can't change your queue.

Get and install it

The skill file isn't published for download. Ask support in chat for a copy. Then save it in your project as .claude/skills/krevaya-content-ops/SKILL.md, or in ~/.claude/skills/krevaya-content-ops/ to use it in every project. Claude Code uses it when you ask it to manage your Krevaya queue, review drafts, check performance, or plan your week.

No install needed: the built-in prompts

Krevaya's MCP server ships the same review-first discipline as prompts that work in any client that supports them: content_ops, weekly_review, repurpose_source, and plan_week. In Claude Code, type / and pick the Krevaya prompt you want. content_ops also walks through reply suggestions and your engagement inbox.

What it won't do

  • Approve Pinterest or Tumblr posts. Krevaya refuses those approvals from any AI app; you approve each one in Krevaya.
  • Hand work to an AI team. Krevaya's separate AI team is switched off. In the app, Krew is one assistant, Krevaya AI, and the skill does the work itself with the ordinary tools.
  • Retry blindly. When a tool returns an error (a missing permission, no free slot, a plan limit), Claude relays the message instead of repeating the call.

Did this answer your question?