The Reply Queue: five conversations a day, every reply standing on a fact you own

Knowing how to reply well was never the problem. Doing it every day is. You built the expertise, you even built the system, and the daily loop still dies because it runs on discipline you don't have to spare at 9 a.m.

So the reply engine now hunts on purpose, right inside X.

What ships today

Open a Hunt in the BlackOps Chrome extension and send it hunting. One click turns the Hunt's keywords into a live search on X itself, opens it in your browser, and scores the results as they render, surfacing the conversations that actually need what you know. Pick one and the engine drafts on demand, through every gate below, staged straight into X's native composer for your review. That's the targeted hunt, and it's the front door.

Between hunts, the extension keeps watch. Your active Hunts ride along as you scroll, flagging the bonus catches in whatever timeline you're already reading. Deliberate first, serendipity on top.

The Reply Queue in your admin is the ledger behind all of it: your drafts and what each one stands on, the targets found but held back and why, the rules the system has learned from you, and a streak counter that keeps score against your daily goal. Hit your goal and the day is done.

The X API can't send replies, and we wouldn't want it to. Every reply waits for your eyes and ships from your hands. What the engine removes is the grind: the triage, the second-guessing, the blank composer.

Six gates between a keyword match and your send button

Most reply tools optimize for volume, so their failure mode is slop. This engine optimizes for the opposite, and it earned every gate the hard way, live, on real tweets:

1. The quality bar. Candidates are ranked 0 to 100 on reach, velocity, freshness, and author fit. Below your bar, they don't exist.
2. The on-topic check. Keywords find homonyms. A hunt watching "stay consistent" will surface a tweet about buying index funds. A semantic pass checks every candidate against what your Hunt actually wants, and it judges the author too: a newsletter feed broadcasting a paper summary is not a conversation, no matter how on-topic the words are.
3. The verified quote. Every draft is anchored to one fact from your Brain, and the anchor must carry a verbatim quote from a note you actually wrote, matched character for character. A fabricated fact can't produce a real quote. This gate is deterministic.
4. The fit check. A fact can be true and still be the wrong thing to say. If your fact doesn't answer what the tweet actually asked, no draft.
5. The hijack check. A reply that pivots someone's thread to your agenda fails review, no matter how polished it reads.
6. The grounding check. The finished draft gets fact-checked against your own material one last time. Claims your notes don't support get rewritten out, or the draft dies.

What survives shows its work. Under every drafted reply sits a receipt: "Standing on:" followed by the verified fact it was built from. You never send a claim you can't see the source of.

The engine that tells you no

Some candidates clear the quality bar and still can't produce a verified fact. Those aren't discarded and they aren't faked. They land in a "Found, but held back" section with the tweet, the score, and the reason, next to a Draft anyway button. Nothing below the gate is ever drafted automatically. Your judgment can override the machine's, but it takes your click.

A thin day is explained, never mysterious: every no is itemized with its reason, right next to the drafts that made it. In a timeline drowning in AI reply-guy slop, the tool that withholds is the tool people trust.

It learns your judgment, not just your voice

Two learning loops run under the queue. The first is wording: edit any draft inline before you send it, and the queue keeps your version next to the machine's, feeding the same corpus that trains your voice profile. Every edit makes tomorrow's drafts sound more like you.

The second is judgment. When a draft misses in a way an edit can't fix, hit Feedback on the card and say what's wrong in plain words: too salesy, wrong fact for this crowd, more like this one. The system distills your reaction into standing rules, applies them to every future draft, and shows them on the queue page under Learned Reply Rules, where one click removes any rule you disagree with. You're not correcting one reply. You're teaching the system how you decide.

Bind a post and it rides along

Your published posts are the densest store of facts you own, so now they're fuel. Bind a post to a Hunt and the drafting pipeline reads the post body itself: anchor facts come quoted, verbatim-verified, straight from what you published.

And when a draft's fact came from a bound post, the reply may carry that post's exact link. Only then. A reply standing on any other fact never links, because a link without its fact behind it reads like a drive-by plug. Links stay rare on top of that, roughly one in twenty of a Hunt's drafts, and the queue card shows which post each draft is anchored on before you send.

That closes the loop most content tools leave open: publish a post, bind it to a Hunt, and the Hunt finds the conversations your post answers, drafts replies grounded in the post's own claims, and sends readers back to it, one earned reply at a time.

Where it fits

The Reply Queue is the review desk on the reply engine you may already be using: Hunts define what you want to be known for, the extension runs each Hunt as a targeted search on X and keeps watch while you browse, the fact-anchored drafting writes only what your Brain (or your bound posts) can prove, and every draft lands in X's native composer for your review. Your AI assistant can read the day's queue too, over MCP, with get_reply_queue.

It's available now on the Pro plan. Install the Chrome extension, create your first Hunt, and send it hunting. The docs cover the rest.

Don't chase trends. Answer the conversations that need what you know.