kevinbadi/hyperedit/respond-to-dms

Respond to DMs

When to use

• The human asks you to reply to DMs, "check my messages," or keep the inbox at zero.
• A scheduled automation run whose job is message triage.
• Cleaning up: archiving stale or spam conversations.

Do NOT use this skill for post comments (that's respond-to-comments) or reviews (inbox:reviews).

Prerequisites

• creatoros CLI authenticated: creatoros auth:check exits 0. If it fails, stop and tell the human to run creatoros init.
• At least one connected account: creatoros accounts:list --pretty returns accounts. Note each account's _id — every inbox command needs --accountId.
• Optional: a brand-voice note at creatoros/BRAND_VOICE.md. If present, read it before drafting anything.

Steps

1. Confirm auth and accounts:

``sh creatoros auth:check creatoros accounts:list --pretty ``

1. Fetch active conversations (per account or across all):

``sh creatoros inbox:conversations --status active --limit 20 --pretty ` Filter with --accountId <id> or --platform <facebook|instagram|twitter|bluesky|reddit|telegram>. Paginate with --cursor`.

1. Read each conversation before replying — never answer from the preview:

``sh creatoros inbox:messages <conversationId> --accountId <accountId> --pretty ``

1. Triage every conversation into one of four buckets (see Judgment rules):

• REPLY — a genuine question or conversation; draft a reply.
• ARCHIVE — spam, bots, or a finished thread; hide it from the active inbox:

``sh creatoros inbox:update-conversation <conversationId> --accountId <accountId> --status archived ` (Reversible: --status active` un-hides it.)

• ESCALATE — sensitive; do not reply, collect for the human.
• LEAVE — needs nothing yet (e.g. the human's own last message is awaiting their counterpart).

1. Send each reply:

``sh creatoros inbox:send <conversationId> --accountId <accountId> --message "<reply text>" ` Attach media with --mediaUrl <url> (upload first via creatoros media:upload <file>). Then mark the thread read: `sh creatoros inbox:mark-read <conversationId> --accountId <accountId> ``

1. Fixing your own mistakes (only your own sent messages, where the platform supports it):

``sh creatoros inbox:edit-message <conversationId> <messageId> --accountId <accountId> --message "<corrected>" creatoros inbox:delete-message <conversationId> <messageId> --accountId <accountId> ``

1. Report to the human: counts per bucket, every ESCALATE thread quoted with its conversation ID, and the replies you sent.

Judgment rules

• Escalate, never answer, when a DM involves: refunds, billing, or order problems; complaints; legal, medical, or financial questions; press/partnership/sponsorship inquiries; anything romantic, harassing, or involving a minor; requests for the human's personal contact info.
• DMs are private but not consequence-free. Screenshots exist. Nothing in a DM that couldn't survive being posted publicly.
• Never send links, prices, or promises the human hasn't already stated publicly or pre-approved.
• Don't start conversations. inbox:create-conversation is only for when the human explicitly asks you to DM someone specific — never cold outreach on your own initiative.
• Archive is the "hide," not delete. Archiving is reversible; deleting messages is not. Only delete a message you yourself sent in error.
• When unsure which bucket, escalate. A slow reply costs little; a bad DM from the brand account costs trust.

Verification

• Each inbox:send must exit 0 and echo the created message. If it fails, retry once; if it fails again, report it as failed — never silently drop it.
• Spot-check one replied conversation with inbox:messages and confirm your reply appears.
• Archived threads no longer appear in inbox:conversations --status active.
• The final report to the human is part of success — no report, not done.