Activity + Posts lane · 2026-08-07

The DM we send, and asking the agent to redo a reply

Two of the four things you raised. The other two were already ruled by the approved sheet and are built rather than drawn: overruling now reads the registry in the registry's own words, and the Source control gets its second destination. These two are not ruled by anything, so they are drawn instead.

Nothing here is built. Pick a fork on each and it becomes a work order. Both are checked against the code as it runs today, not against a plan.

The DM we open with is in Activity, and you cannot tell

present, indistinguishable

You said you cannot see the DM-opener message or action in Activity. It is there. The problem is that nothing marks it as a DM: it is written as a comment turn with the outcome replied, so it renders as kind Comment, outcome sent, exactly like a public reply. On a live send one comment produces two identical-looking rows, and neither says a private message was involved.

What the list shows today, for ONE comment

Time	Channel	Kind	Details	Outcome
14:32	Instagram	Comment	investor.ae comment is this still available?	sent
14:32	Instagram	Comment	investor.ae comment is this still available?	sent

One of these is the public reply. The other is the private DM we opened. There is no way to tell which, and no way to read what the DM said.

Verified in the code. comment-pipeline.ts:1574 writes the opener's trace with the outcome "replied" and the same caseKey as the comment, so turnKind() answers comment for both. The ACT is recorded properly on the audit spine as dm_opener, which is why the panel's history can show it, but the row itself carries no hint.

Below is the shape I would build, given the sheet's rule that a comment is one row and the panel explains it. The DM stops being a second row and becomes a section of the comment it belongs to, so "what happened to this comment" has one answer.

The list: one comment, one row

Kind	Details	Outcome
Comment	investor.ae comment is this still available?	sent + DM

A quiet marker says a private message went with it. Nothing else changes.

The panel: the DM, in full

What the agent producedposted

Yes, still available! Just messaged you, check your DMs.

The message we sent privatelydelivered

Hi! Thanks for commenting. What are you looking for?

Sent before the public reply, so the inbox is not empty when they look.

The fork

Fold it into the comment: one row, a "+ DM" marker, and its own panel section. One comment stays one row, which is the sheet's whole rule, and the DM text is finally readable. Recommended.
Give it its own row, marked as a DM. Keep two rows but make the second obviously a DM (its own kind and its own outcome word). Truthful about the two sends, at the cost of two rows per comment forever.
Leave it as a second Comment row. Only defensible if you never need to know whether the DM went out, which today's fallback work says you do.

Telling the agent what to change, and letting it rewrite

does not exist

You said you cannot make the agent delete a reply and write another one based on changes you ask for. Correct: every path that rewrites a reply today makes YOU write it. "Replace our reply" opens an empty box for your wording. "Draft a reply now" asks the agent, but only on a comment it skipped, and it takes no instruction from you. There is nothing that says redo this, and here is what was wrong with it.

What overruling a sent reply offers now

Replace our reply becomes sent kept as a correction Deletes what the agent posted and posts your wording instead.

Hide their comment becomes hidden Hides it from everyone but them. Reversible.

Every one of these ends with you typing, hiding, or deleting. The agent is never asked to try again.

The path that is missing

Tell the agent what to fix becomes held kept as a correction Removes the reply, asks the agent for a new one using your note, and brings it back for you to send.

What was wrong with it

Too pushy, and do not quote a price on a public comment. Invite them to DM instead.

Ask the agent again Cancel

your note → agent rewrites → held for you → you send or edit

Why it is worth more than the manual path. Your note is the most valuable correction the system can collect: it says what was wrong in your words, about a specific reply, at the moment you noticed. Today that judgement is spent once, on one retyped reply, and never recorded. Stored as a correction it feeds the periodic prompt review, which is the only loop that stops the same mistake recurring.

The forks — two, and the second is the one that bites

What happens to the public reply while the agent rewrites? Delete it first (the wrong reply stops being public immediately, but the comment sits unanswered for a few seconds) — recommended, or leave it up until the new one is approved (never a gap, but the reply you objected to stays public until you come back).
Does your note reach the agent only, or the instructions too? This reply only, and it is kept as a correction for your periodic review — recommended, and it matches the standing rule that nothing is added to the agent's instructions automatically. The alternative is to fold it into the Situations text, which changes the agent's behaviour everywhere from one annoyed sentence typed about one comment.