First-hour engagement comes from the followers you already have, so this checks whether they are the right ones before a single comment gets written.
We have not measured this one, and for this file the reason matters, because the main claim is uncomfortable enough that a reader deserves to see it stated rather than sold.
The claim is that the first action is not writing better comments. It is checking who currently engages with you. The file gives a procedure for that with real numbers: take your last ten posts, list the first ten accounts to engage with each, keep duplicates because a repeat engager is a repeat vote, classify all hundred rows into buyer, referrer, peer or irrelevant, and read the buyer plus referrer share against three bands, 40 per cent and above, 20 to 39, and under 20. It also states the consequence people miss: commenting well into a mismatched network does not merely waste effort, it entrenches the mismatch, because early engagement from the wrong people is what teaches the platform to show you to more of them.
Two other things in it are concrete. Three permitted comment types, each with a structure and a disqualifier: a specific disagreement with a named claim, an extension that adds an uncovered case, and an example from your own work with a number and a caveat. And a stand-alone test that is operationalised rather than gestured at, including the rule that a comment must contain at least one noun the parent post does not.
Who it is not for. Anyone whose audience is not on a public feed at all. Anyone posting under a brand account with a compliance review, where a specific public disagreement is not an available move. And anyone who needs pipeline this quarter, because the file's own arithmetic says a mismatched network takes months.
Measuring one skill honestly costs about twenty model sessions: five runs with it, five without, on real material, each output graded alone by a session that is not told the other arm exists, against a rubric written by somebody who never saw the skill. We have not spent that on this one yet, so it ships labelled rather than ships silently.
How it would be measured. Tier B, because whether a given comment is worth reading is judgement with no mechanical ground truth. The partial objective spine can be graded first and should be: a network-fit check appears before any writing advice and yields a numeric share against a stated threshold, the plan names a daily comment count inside the stated band, every drafted comment is tagged as exactly one of three permitted types, the stand-alone test is applied and its outcome recorded per comment, and the weekly review is defined on profile visits and follows rather than likes. Material would be ten invented briefs, each carrying an audience statement, a hundred classified engager rows and five posts to respond to. The judgement half needs blind pairwise preference on comments shown without their author to readers who hold the stated job.
The gradeable part is thin and the interesting part is not gradeable, which is why this is Tier B. Whether a network-fit check ran, whether a daily count was named, whether each comment carries a type tag, and whether the weekly review is defined on profile visits rather than likes are all string-level checks a skill arm passes by construction. Whether a comment is worth a stranger's attention has no ground truth, and testing it means comments from the same briefs, shown without their author to readers who hold the stated job, judged pairwise.
The confound is worth declaring too. A strong model asked for a commenting strategy already says something reasonable about being substantive and consistent. The honest expected finding is that only two parts move anything: the network-fit gate running before any writing advice, and the weekly cut, because neither is something a helpful default answer does unprompted.
The rule that decides pass or fail was written down before any run was executed and it does not move afterwards. It is in the method note on the hub, along with the full results table including every skill that was tested and cut.
Stated plainly, because a skill that claims everything is useful for nothing.
name and description in the file's frontmatter, so you can also invoke it by name.The best alternative for most people is twenty minutes a day and no system at all, spent under posts written by people they genuinely want to work with. That produces most of the value here, and the file exists mainly for the case where twenty minutes a day has been spent for six months and produced nothing, which is nearly always a network problem rather than a comment problem.
Commercial tools do the surrounding work better. A scheduler removes the publishing friction. A feed-monitoring tool surfaces the posts each morning, which is the genuinely tedious step. Neither will tell you that your existing followers are the wrong people, and neither will cut an account from your rotation because it produced nothing.
Paid distribution is the honest alternative when the timeline is short. If you need the right people to see you this quarter, buy the attention and run this alongside it, rather than expecting an organic composition shift to arrive on a schedule it cannot keep.
--- name: comment-led-growth-system description: Builds a repeatable daily commenting system for a public social feed, starting with a network-fit check that measures whether the people who already engage with you are the people you are trying to reach, then a target rotation, a timing rule tied to the distribution window, three permitted comment types with structures and disqualifiers, a stand-alone test, and a weekly cut based on profile visits rather than likes. This skill should be used when organic reach is flat despite consistent posting, when a daily commenting habit is about to be started, or when engagement is high and none of it converts into profile visits. --- # Comment-led growth system ## The claim this skill is built on The usual approach is to write better posts. When that fails, people write more of them, then add a commenting habit aimed at whichever accounts are largest in the niche. The claim is that this ordering is wrong, and the reason is mechanical. On a ranked feed, a new post is first shown to a slice of the people who already follow you. What that slice does in the first stretch of the post's life is the main input into whether it is shown to anybody else. So your existing network is not the audience, it is the gate, and its composition caps everything downstream. If the people who already follow you are peers, competitors and people who followed you back out of politeness, then a better post buys you more engagement from peers, competitors and polite people, and the post still does not reach a buyer. There is a second-order effect that makes this a gate rather than a tuning step. Engagement from the wrong people is a signal, and the system reads it. Comment well into a mismatched network and you do not stay level, you get worse, because you have taught the ranking system that this is who you are for. That is why the network-fit check is action one and not an optimisation to run once the habit is established. ## Dated mechanics, and how to check them Everything below that depends on platform behaviour is stated as of August 2026 and none of it is published by the platforms. Verify each against your own numbers before relying on it. - **The distribution window.** Roughly the first 60 to 90 minutes on a professional feed. Your own number is visible: watch when your own posts stop accruing impressions, and use that instead. - **Comment length ceilings.** Around 1,250 characters for a comment on the main professional network, 280 characters for a reply on short-form networks without a paid tier. Check in the composer, because these change. - **Reply volume as a signal.** Replies on your own posts appear to feed distribution alongside other engagement. Treat this as a mechanism you can test, not a rule. - **Rate limits on comments and connection requests exist and are unpublished.** A sudden unexplained drop in your own reach is a plausible sign you have hit one. Compose comments in a plain text editor, Notepad on Windows or a text editor in plain text mode on macOS, rather than a word processor. Word processors convert straight quotes to curly ones and hyphen pairs into dashes, and some composers render those as literal artefacts. ## Step 0. The network-fit check, which runs before anything else **The procedure.** Take your last ten posts. If you have fewer than ten, use every post from the last 90 days. For each post, list the first ten accounts to engage with it. That is a hundred rows, duplicates kept, because a repeat engager is a repeat vote. Classify every row into one of four buckets: - **A. Buyer.** Somebody in the audience you have stated, with the authority to act. - **B. Referrer.** A practitioner adjacent to the buyer who can pass your name along. - **C. Peer.** A competitor, a peer, or somebody who sells the same thing you sell. - **D. Irrelevant.** Everybody else, including reciprocal-engagement accounts and general commentary accounts. Compute A plus B as a share of one hundred. **The bands.** - **40 per cent or more: fit.** Proceed to the rotation. - **20 to 39 per cent: partial.** Proceed, but build the entire target rotation from where the A rows came from, and do not add accounts on any other basis until the share moves. - **Under 20 per cent: mismatched.** Stop. No commenting system will fix this. The correct first action is changing who follows you. - **You cannot tell.** Two causes, two different answers. If your posts do not have ten engagers each, run the same classification on your last 200 followers instead and treat the result as directional. If you cannot classify a row because your audience statement is a category rather than a role, that is the actual blocker: write the audience as a job title plus a situation plus who signs off, in one sentence, then re-run. If you cannot write that sentence, no commenting system helps, and the work in front of you is positioning. ## What to do when the check fails Unfollowing people does not change who sees your posts, so that is not the move. Five actions that do: 1. Comment exclusively into the audience you want for 30 days, without publishing for reach. 2. Accept and send connection requests only inside the audience definition. 3. Leave every reciprocal engagement arrangement. Those accounts are the largest single source of bucket D and they are indistinguishable from success in a likes count. 4. Rewrite the profile so somebody arriving from a comment can tell in one line whether you are for them. 5. Re-run the check at day 30, and expect a slow number. The arithmetic is the part worth internalising. Follower composition changes at roughly the rate you add people. Add 100 a month to a network of 3,000 and the composition moves by about three percentage points a month, even when every one of the hundred is perfectly targeted. Moving from 25 per cent to 40 is therefore a four to six month job at that rate. Anybody promising it faster is proposing to buy the audience, which is a legitimate choice and a different one. ## Step 1. The objective and the single conversion surface The output of this system is profile visits from the right people. Not likes, not comment counts, not impressions. That means the profile is the call to action, and it has to carry three things: what you do and for whom, in the first line; something in the opening lines of the summary that survives truncation, since most profile previews cut after two or three lines on a phone; and one link. Fix that before the rotation starts, because every comment sends people there. ## Step 2. The rotation Build a list of 15 to 25 accounts, and comment on 3 to 5 posts a day from it. Selection rule: an account earns a place because its audience is your audience, not because it is large. The reliable way to find them, when the network-fit check produced bucket A rows, is to look at whose posts those specific people comment on. That is a list of accounts your buyers already read, which is a completely different list to the accounts with the biggest follower counts in your category. Cap the daily count. Three to five substantive comments is a sustainable rate that leaves room for the reply pass. Fifteen a day is a pattern that readers and automated systems both recognise, and it costs the credibility the system is trying to build. ## Step 3. Timing Comment while the post is still being distributed. A comment posted a day later is read by the author and nobody else, and it earns nothing regardless of how good it is. This is the single largest source of wasted effort in the whole system, and it is invisible from the inside, because the comment looks exactly the same to you. Practically: record when your rotation accounts publish, over a week. Most people post on a schedule they never change. Then take two 20-minute blocks a day that overlap those times. Blocks beat notifications: a notification-driven session runs an hour and produces four comments, a timed block runs twenty minutes and produces five. ## Step 4. The three permitted comment types Only three things are worth writing. Everything else is a reaction. **1. The specific disagreement.** Name the exact claim you are disagreeing with, state the condition under which it fails, give the reason. Disagree with a stated claim, never with the person. Never disagree with a post that carries no claim: a personal story has nothing to disagree with, so the attempt reads as an attack. **2. The extension.** Accept the post's frame in half a sentence, add the case it did not cover, then say what changes in that case. The test is whether the addition would have belonged in the post. **3. The concrete example from your own work.** The situation, what you did, the number, and the caveat. The caveat is what makes it credible and is the part everybody drops. **Not comments, in any circumstances:** agreement, congratulation, a restatement of the post in your own words, a compliment attached to a question, tagging somebody, an emoji, and the announcement that you are adding your thoughts below, followed by no thoughts. Restatement deserves its own warning because it is the default shape of machine-written comments and it is now widely recognised as one: open by summarising the post's thesis, add a hedged "I would add", produce a three-item list, close with a rhetorical question. If your comment has that skeleton, rewrite it into one of the three types or skip the post. ## Step 5. The stand-alone test and the length band **The test.** Cover the parent post. If the comment still tells a stranger something, it is a contribution. If it becomes meaningless, it is a reaction, and reactions cost you time and earn nothing. Two operational versions of the same test, because the abstract version is easy to pass in your own head. The comment must contain at least one noun that does not appear in the parent post. And it must not open with a pronoun pointing back at the post: "this is so true" and "exactly this" both fail at the first three words. **The band.** Roughly 40 to 120 words. Under about 25 words it is nearly always a reaction. Above roughly 150 it competes with the post it sits under, and gets read by fewer people than a shorter one would. ## Step 6. The reply pass Reply to every reply on your own posts, inside the same window. Each reply is itself a comment and has to pass the same stand-alone test, which is exactly why "thanks!" is a cost rather than a courtesy: it adds nothing to read and it uses up the attention of somebody who came back. ## Step 7. The weekly cut The metric is profile visits and new followers. Not likes on your comments. A comment with 40 likes from the wrong network is a cost that looks like a result. Here is the honest problem: as of August 2026, no major platform attributes a profile visit to the comment that caused it. So direct attribution is unavailable, and the workaround is cohorting plus proxies. **Cohorting.** Run a fixed rotation for two weeks. Record weekly profile views and new followers. Swap out the bottom third, keep everything else constant, and see whether the number holds or falls. **Per-account proxies**, scored weekly, one point each: 1. Did the author reply to you? 2. Did anyone other than the author reply to your comment? 3. Did you gain a follow or a connection request within 24 hours of commenting there? **The cut.** Any account scoring zero for two consecutive weeks leaves the rotation, regardless of its follower count. Without this rule the rotation drifts towards the largest accounts within about a month, because commenting under a big account feels productive in a way that shows up nowhere in the numbers. ## Decision rule: does this post deserve a comment? - **It makes a checkable claim you have direct experience contradicting or qualifying:** comment, type 1. - **It is correct and incomplete in a case you have handled:** comment, type 2. - **It describes a situation where you have a number:** comment, type 3. - **It is a personal announcement, a job change, a bereavement or a celebration:** not part of the system. Say something human if you mean it, and do not count it. - **It is outside the distribution window, or already carries several hundred comments:** skip. Your comment will not be read whatever it says. - **You cannot tell whether you have anything to add:** apply the deletion test. Write the comment, delete the first sentence, and see whether anything survives. If nothing does, you had a reaction. Skip the post and take the next one on the rotation. Do not force one of the three types onto a post that does not support it, because a manufactured disagreement is the most visible failure this system produces. ## Worked example A freelance data engineer selling to heads of analytics at software companies of 50 to 300 people. **Network-fit check.** Last ten posts, hundred first-engager rows. A, heads of analytics and their managers: 11. B, analytics engineers and analysts: 18. C, other freelance data engineers and agencies: 44. D, recruiters, reciprocal-engagement accounts and general commentary accounts: 27. A plus B is 29 per cent, which is the partial band. That is a diagnosis nobody would have reached from the engagement numbers, which looked healthy. Forty-four per cent of the early engagement is competitors, and each of those posts was therefore being distributed further into a pool of people who will never buy anything. **Rotation.** The 11 A rows came from four posts, all four about a specific migration failure with a number in it. The rotation is built from the accounts those 11 people themselves comment under, which turns out to include two vendor engineering blogs' authors and several analytics leads, and to exclude every large general-technology account currently in the habit. **Operation.** Four comments a day, two 20-minute blocks timed to when the rotation publishes, plus the reply pass. Every comment tagged as type 1, 2 or 3 before posting, and the deletion test applied to anything that felt thin. **Week three cut.** Of 18 accounts, six scored zero twice and left, including the two largest in the rotation. They were replaced from the same source: accounts the A rows engage with. **Verdict: this is a partial-fit build with a composition target, not a comment-quality problem.** Four comments a day into a rotation of 18, with the explicit objective of moving A plus B from 29 per cent towards 40. At roughly three percentage points a month for a network this size, that is a four to six month job, and no improvement in comment quality shortens it. What comment quality decides is whether the people who do arrive stay. ## Failure modes **The mismatched-network flywheel.** Comments are good, engagement rises, reach falls or flattens. The early engagement is coming from peers, the ranking system learns the audience, and every successful week deepens the problem. It looks like progress in every visible metric except profile visits. **Big-account chasing.** The rotation fills with the largest accounts in the category, because commenting under them feels significant. Their audiences are broad and mostly not yours, and the comment sits in position 214. **The late comment.** Posted outside the distribution window, read by the author only. Indistinguishable from a good comment from where you are sitting, which is why it survives for months. **The restatement comment.** Summarises the post, adds a hedge, lists three things, ends with a rhetorical question. Reads as machine-written to anybody who has seen a few, and attaches that impression to your name rather than to a draft. **The compliment with a question attached.** Feels like engagement, asks the author to do free work in public, and fails the stand-alone test in the first clause. **The pitch in the comment.** Converts a contribution into an advertisement under somebody else's post, which is the one place readers reliably punish it. The profile is the call to action, and it works precisely because the comment did not ask for anything. **Disagreeing with the person rather than the claim.** The comment stops being about a condition under which something fails and becomes about who is wrong. It gets attention, and none of it is the kind you wanted. **The bare reply.** "Thanks!" under every reply on your own post. It costs the reader nothing to skip and costs you the one moment somebody came back. **The uncut rotation.** The weekly review is skipped three times, the list becomes whatever is convenient, and effort keeps flowing to accounts that have produced nothing for two months. **Reciprocal engagement arrangements.** They manufacture early engagement from a fixed group who are not your buyers, which is the exact input the network-fit check is trying to clean up, and they make the resulting numbers look better while making the outcome worse. ## What this skill does not do - It does not write your posts. It decides whose posts you appear under and what a comment must contain, and it assumes something worth reading exists at the other end of the profile visit. - It cannot see your analytics, so the network-fit check runs on whatever you can classify by hand. That is the input, and the quality of the output follows it. - It has no access to any ranking system. Every distribution claim in it is an inferred mechanism, dated to August 2026, and needs checking against your own impression curves. - It does not measure conversion. Profile visits are the objective here because they are the last thing this system can influence, and what happens after that belongs to your profile, your offer and your pipeline. - It cannot make you credible. The three comment types require a real disagreement, a real uncovered case, or a real number from your own work, and where none of those exist the correct output is fewer comments rather than manufactured ones. - It is not fast, and it says so with arithmetic rather than a caveat. If the audience has to exist this quarter, this is the wrong instrument.
These skills all ask your assistant to check things against your actual codebase, your actual schema, your actual design system. Locul keeps that context current on its own, from the files you already have, on your machine. Mac and Windows, free to start.