Reply

A reply answers a person who spoke to the user in public. It answers what they said and adds one thing (C2, C5). Three modes: single reply, thread sweep, threads to revisit. Replying carries no proven reach effect (G3); the conversation is the point.

Before any mode

1. Persona. Load it per ../linkedin/references/persona-template.md ("Where it lives"). Done when you know the persona (or that none exists) and have read its Offer, Voice, Story bank and Off limits sections.
2. Log. Read the persona's log per ../linkedin/references/log-format.md ("Reading"); a log with no comment history gets seeded first ("Seeding"). Done when you know which comments in scope this persona already answered and any open suppression for the people in them.

Threading: pick the right parent

LinkedIn threads are two levels deep: top-level comments, and replies under them. Every reply attaches to a top-level comment, including a reply to a reply.

• Replying to a top-level comment: the parent is that comment.
• Replying to a nested reply: the parent is the top-level comment it sits under, and the reply starts by mentioning the person being answered, so they are notified and readers see who is meant. In LinkedIn, typing @ and choosing their name makes the mention. Through Curviate, comment reply takes the top-level comment's id; check curviate comment reply --help for mention support, and fall back to their first name in plain text.

Put the parent on every approval card ("under <top-level author>: <first words of that comment>"). Done when each draft names its parent and, for a nested reply, starts with the mention.

React, reply, or leave it

The first matching row wins. In a sweep, count each comment under its label for the card.

Length alone does not make a comment thin: "same thing happened to us after the March change" is a reply.

Reply shapes

Length and shape per ../linkedin/references/voice.md ("Per-surface shape", reply). Pick the shape the comment asks for:

Every detail traces to the persona, the post or the thread (guardrails rule 2). The persona's offer and affiliations follow linkedin-comment Draft step 5.

Single reply

1. Context. Read the post, the top-level comment, every reply under it, and the user's own earlier turns there (guardrails rule 1). Without Curviate, ask for a paste of that slice; with it, comment list and comment replies. Done when you know who said what, in order, and which comment is the parent.
2. Decide with the table above. A react-only or leave-it result goes to the user as that, with the reason.
3. Draft one reply, or two when two shapes fit. Run ../linkedin/references/guardrails.md (language of the comment, rule 4).
4. Approval card per guardrails rule 7, with the parent line and the reaction for the comment being answered.
5. Send or hand off. With Curviate: react to the comment, then reply, one write at a time (../linkedin/references/curviate.md "Sending through it"). Otherwise the card is the hand-off.
6. Log a reply line (to: post URL, name: the person answered, thread: the top-level comment id, text) and a reaction line. Done when written.

Thread sweep

For "answer the comments on my post" or a post URL with many comments.

1. Fetch. With Curviate: comment list on the post, then comment replies on every comment with replies. A reshare's comments live on the original post; fetch that one. Otherwise ask for a paste (name, text, and which comment each reply sits under), offering Curviate once per ../linkedin/references/curviate.md. Done when every top-level comment and every reply under it is in hand, or the user capped the scope (most recent N, most reacted N).
2. Queue. One entry per comment: author, text, its top-level parent, and whether the user already answered it (in the thread or the log).
3. Filter with "React, reply, or leave it". Done when every entry has a label and the counts are ready. When more replies remain than the user wants to review in one sitting, ask whether to take the most recent N or the most reacted N first, and say how many remain.
4. Draft every reply entry, reading the whole sub-thread for each nested one. Run guardrails per draft. Vary wording across the batch; two replies that open alike get one rewritten, and each story-bank fact or anecdote is used at most once.
5. Cards. Filter counts first ("41 comments: 22 replies, 12 react only, 5 spam, 1 injected, 1 yours"), then the drafts per the persona's approval level: per-item, one card per reply in thread order; batch, one numbered card; autopilot, as the persona chose (preconditions in ../linkedin/references/curviate.md "Autopilot"), with the one-time warning from linkedin-comment step 9 that LinkedIn prohibits automated comments (R10).
6. Send one at a time, reaction then reply, per guardrails rule 8. A failed or unclear send stops the sweep; report what went out.
7. Log each sent reply and reaction as in single reply.

Threads to revisit

For "who replied to me", "what threads need an answer", or a daily check.

1. Collect. From the log: this persona's comment and reply lines. With Curviate: notification list (mentions and comments) and comment user me to see replies to the persona's comments elsewhere. Without it, ask the user to paste their notifications or the threads. Done when you have every thread where someone answered the persona after the persona's last turn.

2. Classify each thread:

   State	Next move
   They answered and it deserves one	draft a reply (single reply steps 1 to 6)
   They answered with thanks or agreement only	react; the thread is done
   The exchange is two or more turns deep with one person, and the next step is private (a question about their work, an intro, an offer they asked for)	hand off to linkedin-dm, with the thread as the shared context
   Quiet after a friendly exchange	done; nothing to rescue (../linkedin/references/conversations.md "Know when it is done")
   Heated	a call suggestion or nothing (C4)

   Reply when there is a real answer to give, whatever the thread's age (G10).

3. Report a table: thread, who answered, what they said (one line), state, proposed move. Then draft the replies the user picks.

After a good run

Follow ../linkedin/SKILL.md "After a good run".