Files
deepseek-harness/packages/core/agent-tool-mode/README.md
T
Yichen Jiang 9eaa9d22a5 feat(tools): let one agent choose its tool presentation, and ship code
Code Mode was a deployment-wide field on the host `tools` row: a
deployment ran every session that way or none. The obvious product
shape — 代码模式 beside 标准/极简/创造 in the preset picker — had
nothing to hang on.

The registry itself cannot move into a preset; the agent loop's
scheduler, the api-proxy's presenters, and every tool plugin are its
consumers. So split the registry from its projection: `presentAs(mode)`
writes one cell on the calling agent's scope layer, exactly as
`restrict()` does, and the three reads that decided presentation take
that scope's mode instead of the service's. The config `mode` becomes
the default agents shadow rather than a process-wide fact.

Two consequences are load-bearing. `run_code` now enters a view only
for scopes whose own mode presents it — a native agent must not find it
dispatchable because another agent in the process does — and the
reserved name holds whatever the configured mode, since any agent may
select a code mode later.

`dsh-agent-tool-mode` is the row a preset carries to declare this. A
code mode waits for the host's `codeRuntime` rather than assuming it,
so a runtime-less deployment fails the preset at mount, naming the
row, instead of at the session's first request.

The shipped `code` preset is `standard` plus that row, ordered second.
2026-08-07 00:44:49 +08:00

2.4 KiB

dsh-agent-tool-mode

English | 中文

The row an agent preset carries to say which form of its tools the model sees: native (every schema), code (only run_code plus a generated TypeScript SDK), or both.

Why a row rather than a registry

The tool registry cannot move into a preset. Its consumers are all host-plane — dsh-agent-loop reads its scheduler, dsh-apiproxy reads its presenters to render tool cards, and every tool plugin registers into it — and a service only moves down when all of its consumers move with it.

What a preset can own is the presentation of that registry. ctx.tools.presentAs() declares it for the mounting agent alone, so a Code Mode session runs beside native ones in one process, each seeing its own catalog. The deployment's mode on the dsh-tools row remains the default that agents declaring nothing get.

What it does

native applies immediately. A code mode instead waits for ctx.codeRuntime, which is a host-plane service (dsh-code-runtime-worker): a preset selecting Code Mode against a deployment composing no runtime then holds this row pending, and dsh-agent-presets refuses the mount naming this id. The alternative — applying optimistically — moves the failure to the session's first request, where the operator can act on neither the preset nor the composition.

mode is required rather than defaulted, because a preset without this row already gets the deployment default; an omitted value would mean the row was composed for nothing.

One agent declares one presentation. A second declaration in the same composition is refused rather than merged: two answers to "which form does the model see" is a contradiction, not an override.

Model Experience

Indirectly, through the projection it selects in dsh-tools: code presents run_code plus a generated SDK section, native presents every tool schema.

KV Cache effect

No direct invalidation; the presentation is fixed when the agent is composed, so its request prefix is stable for the session's life.

Known Limitations and Deferred Work

  • The runtime stays host-plane — a preset can select Code Mode but cannot supply the TypeScript runtime it needs; a deployment that composes none can compose no code-mode preset.