[doc] dsh-tools: clarify ToolOutputDefinition.render(args, value) — the SECOND param is the rendered result

[RGarvel]

Summary

The render projection's parameter order is easy to get wrong for plain-JavaScript plugins, which get no TypeScript checking. The signature is render(args, value), but neither the doc comment nor any example states which parameter is "the thing shown to the model". A plugin author naturally writes render = (value) => JSON.stringify(value) — which silently binds value to the tool's input arguments and echoes those back to the model instead of the tool's result. The tool's side effects still run and unit tests often pass, so this only surfaces in a live session (we burned hours on exactly this in a real plugin).

Where

• @deepseek-ai/dsh-tools/lib/types/index.d.ts — ToolOutputDefinition.render
• @deepseek-ai/dsh-tools/lib/types/schema.d.ts — DefineToolOptions.output.render

Current:

/** Pure projection from validated arguments and value to Native/model content. */
render(args: unknown, value: JsonValue): ContentBlock[];

Ask

Tighten the doc comment so the semantics are unambiguous, e.g.:

/**
 * Project the tool RESULT into model-visible content blocks.
 * @param args - the validated tool INPUT (same shape the handler received); usually unused for rendering.
 * @param value - the validated tool RESULT — this is the content to render.
 */
render(args: unknown, value: JsonValue): ContentBlock[];

Ideally add the same note to the defineTool docs/example with a one-line counterexample: "don't write render: (value) => … — that binds value to args."