ryoitabashi/shelly/social-post-drafting

An agent that ends in a Social Post (Discord/Slack/Telegram/Mastodon/ Misskey/WordPress/Bluesky/X connector) action is one step away from publishing something publicly and irreversibly. Draft the post itself as a distinct step, not as a byproduct of the run's raw output:

1. Write for the platform, not for yourself. A Discord/Slack update can

use more structure (headers, bullets); X/Bluesky/Mastodon need a single tight thought within the platform's length norms; a WordPress post is a different register entirely (a title + body, not a one-liner). Don't reuse one draft verbatim across connectors.

1. Strip anything that was scratch reasoning, not the intended message.

The run's internal notes ("let me check...", tool call summaries, intermediate drafts) must never leak into the published text — reread the draft as a stranger would see it before finalizing.

1. **Never include a secret, token, internal path, or another person's

private information** in a public post, even if it appeared incidentally in the source data the agent was summarizing.

1. Respect the one-agent-per-connector limit: don't try to fan the same

content out to multiple platforms from a single Social Post agent — each platform needing the same content means a separate registered agent per connector today.

1. **When the content is ambiguous, ironic, or could be misread out of

context** (sarcasm, a quote, a partial fact), either drop it from the post or make the framing unambiguous — a public post has no space to add "I meant this differently" afterward.