agentleFS
postzen-devpostzen-dev/instagram-agent-skill/skills/ig-dm/SKILL.md

ig-dm

Skill34 starsChanged 29 days agoMITView on GitHub

Write Instagram DMs that get replies - the keyword delivery (and the comment-to-DM automation that sends it through PostZen), the first message to someone who engaged, the collab pitch, and the two follow-ups. Use when the user says "DM this person", "what do I send them", "set up the keyword DM", "outreach message", "how do I follow up", "pitch this brand", or is reaching out to someone specific.

8 other projects use a version of this, 7 of them edited.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

12 rules. 7 are common, 4 are distinctive, and 1 is very specific to this project.

  • 7

    Common

    Most of the 7 edited versions we read keep these.

  • 4

    Distinctive

    This author's own choices. Worth a look.

  • 1

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Common:

    Ask in one batched question:

    line 38
  2. Common:

    If there is no trigger, there is no message.

    line 48
  3. Distinctive:

    If the PostZen MCP tools are in this session, the keyword delivery does not have to be sent by hand.

    line 75
  4. Distinctive:

    If the person has already written to the account (a story reply, a question, an answer to the keyword DM), the thread exists, and with …

    line 190
  5. Distinctive:

    Only the newest 20 or so messages come back with full text.

    line 200
  6. Very specific:

    Write the reply the same way as any DM here: their words, something given, one small ask.

    line 202
  7. Common:

    +4 days: - add something new. Never "just bumping this".

    line 250
  8. Distinctive:

    Follow-ups are sent by hand even with PostZen connected.

    line 258
  9. Common:

    Never automate outreach DMs, and never use a tool that sends on a schedule to people who did not interact.

    line 264
  10. Common:

    Never fabricate having watched something, a mutual, or a shared anything.

    line 267
  11. Common:

    Never open with "Hey! Quick question" and then not ask a question.

    line 268
  12. Common:

    Never send the pitch in the same message as the compliment.

    line 269

“Common” is measured: kept by at least 6 in 10 of the 7 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

ig-dm

Instagram DMs are the only place on the platform where money actually changes hands, and they are also where most accounts burn the goodwill their content earned. The difference is entirely about who moved first.

The three kinds of DM, and only three are worth writing

1. The reply to a hand raised. They commented the keyword, answered the poll, replied to a story, or saved and asked. They moved first. This is 90% of the DMs worth sending and it converts because it is not outreach.

2. The warm approach. Someone whose posts the user has genuinely been commenting on for weeks. There is a shared thread of conversation already.

3. The collab or brand pitch. A specific proposal to a specific account, with a reason it is them.

Everything else is cold DMing strangers, which is what everybody else does, and it is why reply rates sit where they do. If the user is asking for a cold sequence, say plainly that it is the lowest-yield thing they could do with the same hour, and offer the alternative: comment on those ten accounts for two weeks first. Then write it if they still want it.

Before writing, get the specifics

Ask in one batched question:

  1. Who - handle, what they do, and what they posted or did that started this.
  2. The trigger - the actual reason to message today. A keyword they commented, a story they replied to, a post they published. Not "they fit the ICP".
  3. What the user wants - a conversation, a sale, a collab, a referral. Be honest internally, even if the message does not lead with it.

If there is no trigger, there is no message. Say so.

The keyword delivery

The most common DM in this pack and the easiest to ruin. They commented one word. They are expecting the thing. So:

{their name}, here it is: {the thing, or the link}.

{one line on how to use it}

{one question they can answer in four words}

Send the thing first, in message one, with no gate. A keyword post that delivers a "before I send it, can I ask what you do?" is a bait and switch and it is remembered. The question at the end is what starts the conversation, and it is optional for them.

Automated keyword replies are a supported feature for professional accounts, through Instagram's own tools or an approved partner. Using that is fine. Sending unsolicited bulk DMs is not, and it is the fastest route to a restricted account.

Making it an automation through PostZen

If the PostZen MCP tools are in this session, the keyword delivery does not have to be sent by hand. createCommentAutomation watches the comments on a post (or the whole account) and sends the DM for each matching comment, using Meta's one private reply per comment. Write the message as above, then build the call. The parts:

Where it listens.

  • accountId from listAccounts({ platform: "instagram" }). Fixed after creation.
  • name, 1 to 120 characters, for the user's own list.
  • trigger: "comment" (default) or "story_reply".
  • Scope: platformPostId (the Instagram media id) for one post, or postId (a PostZen post id; it resolves when the post publishes, so you can set this up before /ig-publish sends the reel), or neither for the whole account. One active per-post automation per post; a second returns duplicatePostAutomation. A per-post automation wins on its own post; otherwise account-wide ones are tried oldest first.

What it matches.

  • keywords: up to 50, each 1 to 100 characters. Case- and accent-insensitive. An empty list matches every comment, which is almost never what the user wants on a public post.
  • matchMode: "contains" (default), "exact" or "word". For a one-word keyword like CONTRACT use "word", so "contractor" does not fire.
  • typoTolerance: true works with "word" only, and forgives one edit on keywords of 4 to 7 characters, two on 8 or more. It cuts both ways: two edits turn "contract" back into "contractor", so leave it off when the keyword has a longer everyday form.
  • excludeKeywords: up to 50, same mode, vetoes a match.

What it sends.

  • dmMessage: the delivery, written as above. At most 1,000 UTF-8 bytes, not characters: emoji and accents cost several bytes each. With buttons the cap is 640 characters.
  • dmMessageVariations: up to 5 alternates, same limits, picked at random with the base message. Same warmth, different words, so forty people do not post identical screenshots.
  • buttons: up to 3 of { type: "url", title (1 to 20 chars), url }. A link as a button beats a link in the text. Cannot be combined with template.
  • template: an image card instead of buttons. { type: "generic", imageAspectRatio: "horizontal" | "square", elements: [{ title (≤80), subtitle? (≤80), imageUrl, buttons? (≤3) }] }, 1 to 10 elements; several elements render as a swipeable carousel. imageUrl is a public HTTPS URL Meta can fetch. A createMediaPresign publicUrl works and PostZen keeps it for as long as an automation uses it, but upload the file to the uploadUrl before you create the automation, or the card has nothing to show. Any stable public image URL works too. If Meta rejects a card or buttons, PostZen falls back to plain text with the titles and URLs and the log shows buttonsDropped: true.
  • dmDelaySeconds: 0 to 86400, default 0. A short delay reads less like a bot. Thirty to ninety seconds is plenty.

The public reply.

  • commentReply: 1 to 1,000 characters, posted under their comment after the DM succeeds. "Sent, check your DMs" is the whole job. Ignored for story_reply.
  • commentReplyVariations: up to 5, rotated independently of the DM. One base reply plus a few variations is enough; PostZen does not require a minimum.
  • commentReplyDelaySeconds: the reply never lands before the DM, whatever you set.

Who gets it.

  • audience: { followerStatus: "any" | "follower" | "non_follower", minFollowerCount, whenUnknown: "send" | "skip" | "verify" }. Any audience rule forces an opening DM, because Instagram only reveals the follow relationship after the person has messaged the account, and tapping the opening DM's button counts as that message.
  • openingDm: { message (≤640), buttonLabel (≤20) }. Send {} for the defaults: a short "tap below and I'll send you the link" message with a "Send me the link" button. Story replies skip it.
  • followGate: { message (≤640), buttonLabel (≤20), notFollowingMessage (≤1000) }. "Follow to get it" gates. They convert fewer people and they are the bait-and-switch this section warns about, so offer it only if the user asks.
  • Each person gets at most one DM per automation. A second comment logs skipped with already_sent_to_contact. Blocked and unsubscribed contacts (updateContact with isBlocked: true or isSubscribed: false) are skipped too. The account's own comments never trigger.

The rules underneath, which are Meta's: one private reply per comment, ever, and within 7 days of the comment. PostZen caps private replies at 750 an hour per Instagram account and holds the overflow rather than letting Meta reject it. The account needs both the comment and messaging permissions; accounts connected since 2026-09-12 have them. An older one that returns platformCapabilityMissing needs a reconnect. At most 100 automations per user.

The gate. Show the user the keyword, the match mode, the full DM text with its byte count, the buttons or card, the public reply, the scope (which post, or account-wide) and the account handle. Get an explicit yes. Then create it. Then getCommentAutomation({ automationId }) to read it back, and listCommentAutomationLogs({ automationId }) later to see sent, failed, skipped and gated with their reasons. updateCommentAutomation({ automationId, isActive: false }) pauses it; deleteCommentAutomation removes it and its logs. Both are the user's call.

An example, for the CONTRACT reel:

{
  "accountId": "<_id from listAccounts>",
  "name": "CONTRACT clause - Oct 7 reel",
  "postId": "<PostZen post id from /ig-publish>",
  "keywords": ["contract"],
  "matchMode": "word",
  "dmMessage": "Here it is: https://example.com/clause\n\nDrop it in section 4, under payment terms.\n\nWhat kind of work do you mostly contract for?",
  "dmDelaySeconds": 45,
  "commentReply": "Sent, check your DMs.",
  "commentReplyVariations": ["Just sent it over.", "In your inbox now."]
}

Replying to someone who already messaged

If the person has already written to the account (a story reply, a question, an answer to the keyword DM), the thread exists, and with PostZen connected the reply can go out from here instead of from the phone:

  1. Find the thread. listInboxConversations({ platform: "instagram", accountId }) lists them newest first with participantUsername, lastMessage, lastMessageAt and unreadCount. Or searchInboxConversations({ query: "contract", accountId }) to find one by what was said; it searches PostZen's synced copy, matching whole words.
  2. Read it. listInboxConversationMessages({ conversationId, accountId }). Only the newest 20 or so messages come back with full text; older ones may be bare. Reading does not mark the thread read.
  3. Write the reply the same way as any DM here: their words, something given, one small ask. Run it through /ig-human.
  4. The gate: show the handle, the last thing they said, and the reply. Get the yes.
  5. sendInboxMessage({ conversationId, accountId, message }). Set an Idempotency-Key so a retry cannot double-send. For an image, send attachmentUrl plus attachmentType: "image" in a separate call; text and attachment in one call is a 400, and the URL must be public because Meta fetches it. markInboxConversationRead is optional and local to PostZen.

The hard limit is Meta's: a send only works within 24 hours of the person's last message. Outside that it fails with PLATFORM_LIMITATION, and the honest move is to say so rather than hunt for a way around it. A 502 providerOutcomeUnknown means the send may or may not have landed; read the thread again before sending anything. Sends are capped at 120 an hour per user.

What PostZen cannot do: open a new thread. sendInboxMessage takes a conversationId and no recipient, so a person who has never messaged the account cannot be messaged from here. That is why the next two sections stay copy-ready.

The first message to someone warm

  • Two to four sentences. A screen of text is a delete.
  • Reference the specific thing. The comment, the post, the reply. In their words.
  • Give before asking. A number, a template, a name, an answer.
  • One ask, small. "Worth a quick call?" beats "let me walk you through the platform".
  • No link and no calendar in message one. It reads as a funnel because it is one.
  • No voice note to a stranger. It is a great tool and it is for people who already know the user's voice.

The collab pitch

Four lines, in this order: what you have watched them do, the specific idea, what they get, what you need from them. A collab post lands on both grids and reaches both audiences, which is the strongest single growth mechanic on Instagram that does not involve paying anybody. Pitch it as that and be concrete about who does what.

Follow-ups

Two. That is the number.

  • +4 days - add something new. Never "just bumping this". If there is nothing new, there is no follow-up.
  • +10 days - the close-the-loop message. Say you will stop, and mean it. This one gets a surprising share of the total replies, because it removes the pressure.

Then stop. A third converts nobody and costs the relationship.

Follow-ups are sent by hand even with PostZen connected. At +4 and +10 days the 24-hour window has closed unless the person wrote back in between, and if they wrote back, it is a conversation, not a follow-up.

Never

  • Never automate outreach DMs, and never use a tool that sends on a schedule to people who did not interact. It violates Instagram's Terms of Use and restricts the account.
  • Never fabricate having watched something, a mutual, or a shared anything.
  • Never open with "Hey! Quick question" and then not ask a question.
  • Never send the pitch in the same message as the compliment.

Output

The message, the character count, and the two follow-ups with the day each goes out, all run through /ig-human.

Then who sends what:

  • Keyword delivery: a PostZen comment automation if connected and approved, otherwise by hand.
  • Reply in an existing thread, inside 24 hours: sendInboxMessage after the yes, or by hand.
  • The warm first message, the collab pitch, the follow-ups: by hand, always. No API opens a new thread, and that is the right constraint.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.