--- name: log-slack-dm-reply description: "Reading and answering a Slack DM. A log play (docs/useCases/slack/dmReply.md): Read what someone wrote to the account on Slack and answer it in the owner's voice, on the account's own session, with no browser." when_to_use: "Use when the owner asks for this, in words like: \"read the messages with on slack\", \"reply to on slack\", \"what did say in \"" metadata: summary: "Reading and answering a Slack DM" --- # Reading and answering a Slack DM This is the log play `slack/dmReply`. Before anything else, call the log MCP tool `getGuide({ useCase: "slack/dmReply" })`: 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 the owner's real replies (`voice` with no text, `search`), never from memory, and never send two that read alike (`sameAsSent`). Read what someone wrote to the account on Slack and answer it in the owner's voice, on the account's own session, with no browser. How the owner asks for it: "read the messages with on slack", "reply to on slack", "what did say in " ## Run it from a fresh context Say: **"go on docs/useCases/slack/dmReply.md for "**. The agent reads this page and runs the steps. ## The steps | # | Step | Tool | Done when | | --- | --- | --- | --- | | 0 | The account is known: a local row with its `d` cookie | `localAccountList` {}; if missing, the owner runs `log slack cookies save --account ` in a terminal | the account is listed | | 1 | Session check, and which workspaces it is in | `slackSessionCheck`, then `slackWorkspaces` | `loggedIn`, with the workspace's domain | | 2 | Find the chat by who it is with | `slackChats` { account, workspace } | one DM or group with that person | | 3 | Read the chat | `slackRead` { account, workspace, chat, limit } | their latest message and what came before it | | 4 | Pull how the owner writes to them, then draft only from facts the owner gave | `voice` { to, with, context } (no text) | a draft that answers what they asked | | 5 | Gate the draft | `voice` { to, with, context, text } | it passes | | 6 | Send it, one message per line | `slackSendLive` { account, workspace, chat, text, approvedBy: "", confirmLive: true } | Slack answers each line with its ts | | 7 | Tell the owner what they said and what went out | the reply to the owner | both quoted with their times | New messages arrive by themselves: the inbox hook keeps `log slack watch --inbox` open for every signed-in account and puts each new DM, group message or mention in front of the agent (`>>> NEW SLACK MESSAGE`). ## What the runs taught - **Read the chat before sending.** The voice gate inside `slackSendLive` reads the chat's context from the .db, not from Slack (rule 137). `slackRead` stores what it reads there as backfill, so step 3 is what gives the gate the chat's history. Once a thank-you was refused three times while the .db held only 2 of the other person's messages, and the gate read it as a cold first message with an ask in it. Backfilled history is never shown as new messages. - **Send by who it is with.** Give `chat` as the person (their name or handle), not only the D… id, so the gate matches the owner's history with them by name. - **Don't stack messages on a chat the owner already answered.** When the owner's own messages are the newest in the chat (the owner has already said thanks), a third message is refused as not the owner's voice, and it reads like it. Answer it with a reaction instead (`slackReactLive`, [inbox/reactions](../log-inbox-reactions/SKILL.md)), or leave it. A šŸ™ on a contact's short thank-you went out this way.