agentleFS
matthewdonsemail-labmatthewdonsemail-lab/log/.claude/skills/log-twitter-reply-guy-funnel/SKILL.md

log-twitter-reply-guy-funnel

Skill63 starsChanged yesterdayMITView on GitHub

Reply guy, then the DM, then the ask. A log play (docs/useCases/twitter/replyGuyFunnel.md): The owner's way of getting new people to do something for him (share a post, try a tool): like a run of their posts, reply to them in public, then open a DM on their own post, build on what they say, and only then the ask. In his words: \"you're meant to reply guy them first and foremost. You go on, like a tweet, send a meme to the tweet, and then we're gonna forward the tweet to them as an opener, talking and building on top of it, and then you wait for the reply, and then you say something that's in context that then build on top of the conversation that then funnel them into doing an action.\"

The file

We split this skill into the rules it gives an agent

3 rules. About 0 are common, 3 are distinctive, and 0 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈3

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 0

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    following: before the last like, the follow, then the last like a few seconds later, as the contact did.

    line 77
  2. Distinctive:

    Held items: if they message the owner or answer anything mid-run, the run stops liking and the queue holds the next item (rule 109).

    line 79
  3. Distinctive:

    Every send counts against the account's writes for the day, and X's own daily limit (344) can come at around 40.

    line 86

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “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

Reply guy, then the DM, then the ask

This is the log play twitter/replyGuyFunnel. Before anything else, call the log MCP tool getGuide({ useCase: "twitter/replyGuyFunnel" }): it returns this page and makes it the session's play, so every log tool's answer says which step you are on and a send with no play loaded is refused (rules 157, 158). Then follow the steps in order.

Draft every message from that person's own thread and matt's real replies (voice with no text, search), never from memory, and never send two that read alike (sameAsSent).

The owner's way of getting new people to do something for him (share a post, try a tool): like a run of their posts, reply to them in public, then open a DM on their own post, build on what they say, and only then the ask. In his words: "you're meant to reply guy them first and foremost. You go on, like a tweet, send a meme to the tweet, and then we're gonna forward the tweet to them as an opener, talking and building on top of it, and then you wait for the reply, and then you say something that's in context that then build on top of the conversation that then funnel them into doing an action."

It is managerMaxxing.md's touches for anyone: new followers, people who engaged, or a list the owner names. People already talking to the owner skip the first touches and go to getTheAction.md.

How the owner asks for it: "reply guy them first and foremost", "send a meme to the tweet then forward the tweet to them as an opener", "message the people who are new to following me", "get it in front of them", "funnel them into doing an action".

Run it from a fresh context

Say: "go on docs/useCases/twitter/replyGuyFunnel.md for <who> and <the action>" (for example: the new followers, share https://x.com/<account>/status/<postId>). The agent reads this page, README.md and networkMaxing.md, loads getGuide({ topic: "xSending" }) and getGuide({ topic: "extensions" }), and runs the steps.

The steps

#StepToolDone when
1Who: the new followers from the account's notifications, or the people the owner names. Drop anyone already talking to the owner (they go to getTheAction), anyone over 1500 followers, the owner's own accounts and anyone the agent memory says never to messagetwitterNotifications { account }, twitterDms { account, with }a list of new people, each with their user id
2Each person's recent original posts, newest first, the ones with a picture or video marked. Someone with no post from the last month has nothing to reply to: hold them and say sotreg treg.x.user.posts { username }one post per person worth replying to and up to 6 to like, or held
3The like run, timed the way a real like run was: like 3 to 6 of their recent posts, the picture and video posts first, at random gaps of 4 to 60 seconds, the whole run inside 5 minutes and never two at the same gap (rule 94). Follow them a few seconds before the last like. Their notifications fill with the owner's name before the reply lands. The post the reply is on isn't in the run: the reply likes it itselftwitterLike { account, tweetUrl }, twitterLikeLive, twitterFollowLiveeach like answered liked by X, or the run stopped on a refusal
4A reply about their post, with a meme picked to the owner's taste for that moment and uploaded as media. Each reply is about that post alone, so no two read alike (rule 157). Gate every onememelordSearch, voice { to, context, text }every reply passes, and none repeats another
5The replies go into the account's one queue, a few minutes apart, each liking the post and following them first (rule 105)twitterQueueBuild { action: "reply" }, twitterQueueCheck, twitterQueueApprove, twitterQueuePlan, twitterQueueRunLiveX answers sent for each reply
6Once a person's reply has landed, the DM: that same post of theirs forwarded first, then one line that builds on it and on the reply. The opener only, never the ask, written from their post so no two read alikevoice { to, context, text }, twitterQueueAdd { action: "dm" }each opener passes and goes into the same queue
7Wait for them to answer. The inbox hook brings each answer in the moment it lands; read their whole thread thentwitterDms { account, with }they answered, or they haven't yet
8Answer what they said, in the owner's words for them, building on it: about them and what they're doing, still no askvoice { to, with, context, text }, twitterDmLivethe conversation is going
9Once they're talking, the ask in context: the action the owner wants (share the post, try the tool), in a way that fits what they said, with the linkvoice { to, with, context, text }, twitterDmLivethey did it, or the owner knows where it stands

The like run, from a real exchange

A contact once liked a run of the owner's posts. The owner opened the chat by joking about all the notifications, and the contact said they liked to like in bursts. They were talking within the minute. The owner then had the chat distilled into this page: a person who sees the same name across their notifications looks at who it is, so the reply and the DM that follow land on someone who already knows the name.

  • Their posts with a picture or video first. The owner: random intervals "would be good for the media based content the most". A like on a post they made something for gets noticed more than one on a text post.
  • Their timing, guessed from what X shows. X bundled the likes into one notification ("<name> liked 5 of your posts") with one time, so each gap isn't known. What is known: the owner's notification count was 0 about five minutes before, the contact followed, and the last like landed 5 seconds after the follow. So 5 likes and a follow inside 5 minutes, the last like 5 seconds after the follow. The run copies that shape: random gaps of 4 to 60 seconds, all inside 5 minutes, the follow just before the last like.
  • Random gaps, never a burst. Each gap is drawn fresh, never under the governor's 4 s floor and never two the same: likes fired back to back are what X reads as automation and what rule 94 forbids. Each like and the follow is a write the governor counts.
  • Then the reply. It comes after the last like, on a post that wasn't liked in the run (the reply likes its own post).

As a state machine (outlined, not built yet)

Each person is one run of a machine in the account's one queue (rule 105). The queue gains a like action, and the run picks a new random gap before every like.

stateDiagram-v2
  [*] --> picking
  picking --> held: no post in the last month
  picking --> liking: posts found, media first
  liking --> waiting: liked
  waiting --> liking: random gap of 4 to 60 s, likes left
  waiting --> following: one like left
  following --> liking: a few seconds later
  waiting --> replying: run of 3 to 6 done, inside 5 min
  liking --> stopped: X refused
  replying --> opener: reply sent
  replying --> stopped: X refused
  opener --> listening: opener sent
  listening --> talking: they answered
  talking --> asking: the conversation is going
  asking --> [*]
  held --> [*]
  stopped --> [*]
  • picking: their recent original posts from treg, the picture and video posts first, 3 to 6 to like and one to reply to. With no post from the last month the person is held.
  • liking: one like through the account's one session. X's answer decides (rule 73). A refusal stops the whole queue and waits for the owner (rule 94).
  • waiting: a random gap of 4 to 60 seconds, drawn fresh each time, never below the governor's 4 s floor, with the whole run kept inside 5 minutes.
  • following: before the last like, the follow, then the last like a few seconds later, as the contact did. The account's one session is held for the run, so no other person's item goes out in the middle of it.
  • replying, opener, listening, talking, asking: steps 4 to 9 above, unchanged.
  • Held items: if they message the owner or answer anything mid-run, the run stops liking and the queue holds the next item (rule 109).

What to know

  • Never skip the reply. A cold DM with the ask in it to someone who just followed is what X reads as automation, and what the owner never wants. The reply comes first, the opener second, the ask last.
  • No two messages alike. A reply, opener or ask that repeats another one the account sent in the last 24 hours is refused with sameAsSent (rule 157). A template with a different first line is still a template.
  • One play at a time. Loading this page makes it the session's play, and every log tool's answer says which step it is and what comes next.
  • Every send counts against the account's writes for the day, and X's own daily limit (344) can come at around 40; spread the people over days. Any refusal stops the run (rule 94).

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.