AI replies on X, approved in Slack

Company replies with an agent are drafts, not posts. Your agent reads your X mentions and searches for posts in your space, drafts up to three replies in your voice from team memory, checks each one for filler, and posts the drafts to a Slack channel. A person reads every draft and decides what goes out.

The job. Draft replies from the company X account in your voice, and send them to Slack for a person to approve.

Paste this into Claude Code, Codex or Cursor

set up https://corespeed.io/SKILL.md (read it raw, e.g. with curl), then find this hour's X posts that mention our product or our space, draft up to three replies in our voice from team memory, check each for filler, and post the drafts to #marketing in Slack for approval. Do not post to X.

One prompt.Three steps.

Set up first. Save your voice and product facts to shared memory; the agent asks you to connect X and Slack when it needs them. The prompt stops at the Slack draft, because X's automation rules forbid scripting its website.

  1. Teach it your voice

    Save how you talk and what you sell to shared memory. A correction you save reaches every later draft.

  2. Draft, check, send to Slack

    Up to three drafts a round, each checked for filler such as em dashes and lists of three.

  3. Approve, then post

    A person reads each card in Slack. What gets a yes is posted by a person, within X's rules.

The full write-up, on the blog →

The tools it calls

  • memory__search · Read your voice and product facts
  • twitter__get_my_mentions · Read your mentions
  • twitter__search_posts · Find posts in your space
  • slack__post_message · Post each draft for approval
  • slack__get_reactions · Read who approved

Every tool and its price: X (formerly Twitter), Slack

Illustrative example · Fictional data

Everything your agents need, one line away.

Give this to your agent

set up https://corespeed.io/SKILL.md
Get started
FAQ

Questions people ask.