agentleFS
matthewdonsemail-labmatthewdonsemail-lab/log/.claude/skills/log-inbox-triage/SKILL.md

log-inbox-triage

Skill63 starsChanged 2 days agoMITView on GitHub

Triage: scoping someone's job and handing it to whoever does it. A log play (docs/useCases/inbox/triage.md): A person comes to the owner needing something built. The owner is the face of the work: the person on the other side deals with him and nobody else. Behind that, the work goes to whoever actually does it. This page is that route: read what the person needing it has said, get the scope out of them, and hand the whole thing to the person who will build it, as a brief, without ever telling either side about the other.

The file

We split this skill into the rules it gives an agent

1 rule. About 0 are common, 1 is distinctive, and 0 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈1

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 0

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    The doer's replies are the owner's replies: When the client asks how it is going, the answer comes from what the doer has actually reported.

    line 46

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Triage: scoping someone's job and handing it to whoever does it

This is the log play inbox/triage. Before anything else, call the log MCP tool getGuide({ useCase: "inbox/triage" }): 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 matt's real replies (voice with no text, search), never from memory, and never send two that read alike (sameAsSent).

A person comes to the owner needing something built. The owner is the face of the work: the person on the other side deals with him and nobody else. Behind that, the work goes to whoever actually does it. This page is that route: read what the person needing it has said, get the scope out of them, and hand the whole thing to the person who will build it, as a brief, without ever telling either side about the other.

How the owner asks for it: "triage this", "scope it out with him and pass it to <person>", "inform <person> of what the job is and what he needs to do", "we've already scoped the work with the client, hand it over", "this is a triage job", "don't tell him anyone else is doing it"

Run it from a fresh context

Say: "go on docs/useCases/inbox/triage.md for <the person who needs it>". The agent reads this page and runs the steps.

The steps

#StepToolDone when
1Read the person needing it in full, before asking them anything: the whole thread, every file or sheet they sent, what they have already done, what they are stuck on, and what they have said about moneychatsList { account } to find the chat, then messagesRead { account, chat, limit: 120 }their situation in a few lines, with what is done and what is blocked
2Scope the rest out of them, one question at a time: what the thing is for, who uses it, what "done" looks like, what they have already got working, what they are working with, and their budget if it is not knownvoice { to, with, context } then the send, one question per messagethe answers cover what it is, what is done, what is blocked, what done looks like, and the constraints
3Keep the owner as the only person they know. Nothing said to them names, hints at or implies a third party: no "my lad", no "my team", no "someone else will do it", no "I'll get it over to him". They are buying the owner, so the work is the owner'sthe draft itselfthe draft reads as the owner doing the work, and could not be read any other way
4Pick who does it and read their chat: whether they can do this kind of work, what they are owed or have been promised, and what they were last toldmessagesRead { account, chat }the doer, and what they already know about it
5Write the handover as a brief: everything the person needing it said, the whole map of the work, exactly what is blocked, the baseline that already passes, the constraints on their side, and what to send back first. Say what done looks like and that every claim needs evidencethe agent, OUTBOX/<PERSON>_<JOB>.md when it is longit answers clear, why, done when, kind and how to ask (rule 109)
6Claim the doer's chat and send it: one message per line, as a brief, over the gate (a task to the team always goes over it, rule 137)inboxClaim, messagesSendLive { account, chat, text, brief: true, approvedBy: "<owner>", confirmLive: true, overGate: true }the platform answers sent
7Stay the single point of contact on both sides. The person needing it is answered as the owner, from what the doer reports. The doer's questions come to the owner. Nothing is forwarded between them, and no message from one is pasted to the other with a name in itthe two chats, read before every replyboth sides only ever talk to the owner
8Tell the owner where it stands: who has it, what is blocked, what the doer needs, and anything only the owner can answerthe reply to the ownerone line per side

What the runs taught

  • Never tell the person needing it that someone else is doing it. On 2026-11-12 a draft to a client said the work would be passed to "our lad". That single line would have told him he was buying a middleman rather than the person he had been talking to, and it was stopped before it went out. The client's whole reason for paying is that he is dealing with the owner. Keep every handover invisible: the doer gets the map, the client gets the owner.
  • The scope is already in the thread if it was scoped properly. Before asking the client anything again, read what he has already told you: what he has running, what he has tried, what he is stuck on, and any sheet he sent. On 2026-11-12 the whole job was in one status spreadsheet and one explanation of two blocked substeps, and nobody needed to ask him again to hand it on.
  • The handover is the brief, not a summary. The doer needs the client's own map, not a paragraph: the stages, the numbered blockers, the counts that already pass, the constraint that the client's own tooling is out of action, and the first thing to get off him. A short summary produces a doer who starts again from nothing.
  • Say what evidence means before they start. The work here is proving behaviour matches an original, so "done" is a test or a comparison, never an opinion. Say that in the brief or the first report back is a guess.
  • Give the doer the number, the client the number, and never the two together. The doer is paid out of what the client pays. Tell the doer what he is getting; tell the client what the job costs. Neither needs the other's figure, and the person needing it should never learn that the work is being done for less than he is paying.
  • Read the doer's chat before promising anything to the client. A doer who has gone quiet, is owed money, or has said he will not do this kind of work can't be handed the job because the client is keen.
  • A price waits for the owner. What to charge, and whether to move off it, is the owner's call every time. Bring him the client's budget and the doer's number, and let him set the figure.
  • The doer's replies are the owner's replies. When the client asks how it is going, the answer comes from what the doer has actually reported. Don't invent progress, and don't promise a date the doer hasn't given.
  • A play must be loaded before sending as the owner. A send with no play loaded is refused playMissing, and one this page doesn't name is refused playNotCovering. This page is that play.
  • It runs on whichever platform the chat is on. chatsList, messagesRead and messagesSendLive take the platform from the account, so the person needing it can be on one platform and the doer on another. brief is taken on Discord, and overGate on Discord and X. On a platform whose send takes neither, the call is refused inputUnsupported and nothing is sent: tell the owner, who decides whether it goes as an ordinary message through that platform's voice gate.

Next

  • Handing the job over is bossMaxxing.md, which carries the send, the monitor and the walk-through.
  • Working out what a person is worth to the owner in the first place is collabMaxxing.md.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.