Decisions

Why scheduled work and subagents use bounded tools and retained message delivery.

Why doesn't the scheduler ask before it starts a subagent?

Because every subagent would otherwise wait on you, and the caps already bound them. Whether task, schedule and cancel ask first is an opinion, so a key of the agent's spec ([tool.scheduler] ask), and Link's is not to ask. With it on, the three aren't allowed when the tool is defined, so each asks as any tool's call you haven't allowed does.

Can an agent turn its scheduler off?

Yes, in its spec ([tool.scheduler] enabled = false): at the next start the scheduler isn't defined, nothing runs its jobs, and a harness that reads who a message is from has none. Without the key, every such harness has one, since a subagent and a schedule are what people ask an agent for. There's no command for it: the key is the one way, as for every other opinion in the spec.

Why can a spec turn a browser or desktop on?

So an agent can come with its screen. A browser or desktop is defined for each machine, its program's path with it, so an archive can't carry the definition. It can carry the agent's spec, and [tool.browser] enabled = true there makes its first start define one wherever it lands, installing the module as lnk agent tool add browser does. false takes it away at the next start. Link's own spec says neither, so an agent whose spec is silent keeps what lnk agent tool add and remove gave it, and those two commands write the key as they change the tool.

What wakes the scheduler?

The agent's bridge, by Link's spec, since it already runs for the connections and restarts with the machine: the runner sleeps beside it until a job is due. A machine where a resident process costs too much can say [tool.scheduler] driver = "cron" instead: the machine's calendar starts the runner once a minute, and nothing of it runs between.

A key, not a guess

It's a key, as every opinion is, rather than something Link guesses from the machine. A machine that runs an agent only while it answers says address: the scheduler tells it when the jobs are next due, and it wakes the agent then, so even a calendar's minute isn't spent on an agent with nothing due.

The bridge still runs

Whichever wakes the scheduler, the bridge still runs, since a firing reaches the agent through its way in. Where one bridge carries every agent, one runner fires every agent's jobs in a time zone, each through that agent's way in, and an agent's started again only stops once its firings have ended, so none is fired twice.

How does a subagent get fewer tools?

A subagent is a new conversation of the same agent, so by itself it has every tool its parent has. Narrowing it is a list per conversation, in the harness contract (tools): the subagent's task comes with Link-Tools, the names of the tools it may use, and the harness keeps that list with the conversation, offering it only those from then on.

Never widened

A later list is intersected with it, so nothing widens a conversation again. The harness tells its tools which a conversation has (_meta link.local/tools), so the scheduler gives a subagent its parent's tools or fewer, a narrowed parent's subagent included, and never more.

Additive

It's additive: a harness that doesn't say tools gives a subagent all of them, as before, and a list is refused for it rather than ignored, since a subagent with tools it was meant not to have is worse than one not started. The mini harness answers it and the conformance kit checks it.

Link Harness keeps the list beside the conversation's events, not in them: it's a grant, like the agent's settings, and finding it in the events would read every event of every conversation at each turn.