Skip to content

Tinker

Tinker

Knows when you need an agent, a skill, or something else entirely, then builds it.

Tinker carries the judgment call nobody else does: what shape should this take. An agent, a skill, an automation, a small app, or some combination. It reasons that out briefly and in the open, then builds the thing rather than describing it.

It defaults to showing a working first pass. A rough version you can react to beats a flawless proposal you have to imagine.

Call it when

  • You want something built and you are not sure what form it should take.
  • The thing has more than one piece.
  • You want to sit down and make something on purpose, rather than have it offered in passing.
  • You only know that something is slow or annoying. Tinker will ask what you are actually doing repeatedly before jumping to a build.

What it will not do

It will not act without confirming the concrete scope. The trigger is side effect, not size. A draft or preview in the chat is free. Any file write, command, install, credential use, network access, automation or published output gets the specific proposal and target shown first, and waits.

It will not take over Berdy's moment. An obvious singular ask, like automating one weekly thing, does not need to become a build session.

It will not route the other agents. Tinker's instructions name a Conductor agent for that, and no Conductor ships in v0.6.2. Coordinating who works next is yours.

It will not oversell. If what you need is one automation rather than a new agent, it says so. The simplest correct answer, not the most impressive one.

The pattern it will point at

Asked whether something should be an agent or a skill, Tinker often lands on both, and Copycat is the reference case: an agent that is really a thin front end for a skill it creates and maintains. That combination is a good design, not a compromise. The persona holds the conversation; the skill holds the reusable knowledge and can be pulled into any other chat.

Hands off to

Hand to When
Agt. Builder The build needs a real persona, and the voice and boundary deserve a proper conversation
Debugger What got built does not work and the cause is unclear
Pushback The design is worth pressure-testing before it becomes permanent

Handed to it by

Handed from Why
Berdy The need is ambiguous or multi-part. Tinker picks up mid-conversation and confirms only what is missing
Agt. Builder The persona is right, but it needs a supporting skill or tool

Working with it

@Tinker I track reading notes in three different places and keep losing them. Decide whether this should be a skill, an automation, a small tool or some combination, give me the trade-off for my case, then propose the smallest concrete build.

Expect a short reasoning pass, a proposal with the actual target named, and then a pause for your approval before anything is written.

Learn more