deepseek-harness/docs/core-data-structures/tools.md

# Tools

The tool pipeline of [dsh-tools](../../packages/core/tools). [core.md](core.md) introduces `ToolDefinition` as the one pipeline-authoring type promoted to the spine and `ToolSchema` as the model-facing wire shape. This page owns the full `ToolDefinition`, the typed schema DSL that builds it, the guarded execution shapes, and the UI-presentation vocabulary.

Source: [`packages/core/tools/src/index.ts`](../../packages/core/tools/src/index.ts) · [`packages/core/tools/src/schema.ts`](../../packages/core/tools/src/schema.ts) · [`packages/core/tools/src/presentation.ts`](../../packages/core/tools/src/presentation.ts)

## `ToolDefinition` — a registered tool

A `ToolSchema` (the model-facing fields) plus a mandatory canonical output declaration, the `execute` function, host-only scheduler metadata, and optional UI presenters. The registry holds these; the loop dispatches calls through them. The registry's `schemas()` builds the model-facing `ToolSchema[]` by an explicit allowlist — `output`/`execute`/`timeoutMs`/`isConcurrencySafe`/`presentCall`/`presentResult` must never leak into a model request.

```ts type-equiv
/** Tool-owned canonical output contract used after the body returns a JSON value. */
interface ToolOutputDefinition {
  /** Raw supported JSON Schema enforced against every successful canonical value. */
  readonly schema: JsonSchemaNode
  /** Pure projection from validated arguments and value to Native/model content. */
  render(args: unknown, value: JsonValue): ContentBlock[]
  /** Pure replayable presentation projection, computed only for surface calls. */
  presentationMeta?(args: unknown, value: JsonValue): JsonValue
}
```

```ts type-equiv
/** A registered tool: its schema plus the execution function. */
interface ToolDefinition extends ToolSchema {
  /** Mandatory canonical output declaration. */
  readonly output: ToolOutputDefinition
  /** Execute the tool and return only its canonical lossless-JSON value. */
  execute(args: unknown, exec: ToolRunContext): Promise<unknown>
  /**
   * Cooperative tool-call timeout budget in milliseconds. Omit for no deadline.
   * Enforced by `@deepseek-ai/dsh-timeout-policy` (a `tools/execute` wrapper); it
   * is NEVER sent to the model — `schemas()` whitelists only name/description/
   * parameters. Declaring it asserts this tool forwards `exec.signal` to a
   * cooperative implementation that can reach quiescence when the signal aborts.
   */
  timeoutMs?: number
  /**
   * Pure synchronous classifier for overlap with sibling tool calls. Only
   * `true` opts in; omission, exceptions, non-`true` returns, and invalid
   * `defineTool` arguments are exclusive. This metadata is never model-visible.
   *
   * Opted-in executions must not mutate parent-owned state. Shared state must
   * tolerate concurrent dispatch; recorder races are permitted only when they
   * commute or fail closed. See the
   * [parallel-tool-call Agent Note](../../../../.agents/notes/implemented/feature/2026-07-10-parallel-tool-call-execution.md)
   * for the full contract.
   * @param args - parsed arguments; `defineTool` validates before calling.
   * @returns Whether this call may join a parallel group.
   */
  isConcurrencySafe?(args: unknown): boolean
  /**
   * Optional: how to present the PENDING state of one call in a UI, derived from
   * the call's `args` (parsed arguments, `unknown` — the tool validates/narrows
   * its own input). Returns a {@link ToolCallView} (a `card`-tagged render intent),
   * or `undefined` (or omit the method) to fall back to a generic presentation
   * (title = tool name, raw args as input). Pure and side-effect-free: a UI may
   * call it during live streaming AND a session-log replay, so it must depend
   * only on `args`.
   */
  presentCall?(args: unknown): ToolCallView | undefined
  /**
   * Optional: how to present the COMPLETED state, given the same `args` and the
   * durable result projection (`content`, failure state, and optional `meta`). Returns a
   * {@link ToolResultView}, or `undefined` (or omit the method) to keep the
   * pending title and render the raw result content. Pure and side-effect-free
   * for the same replay reason.
   */
  presentResult?(args: unknown, result: ToolResult): ToolResultView | undefined
}
```

`execute` receives `args: unknown` — a raw `ToolDefinition` validates its own input. First-party tools don't write that by hand; they use `defineTool`, which validates and narrows the arguments, infers the body return from `output.schema`, and types both output projectors.

## The unified JSON-value schema DSL

Plugin authors use one vocabulary for typed parameters and typed output values. `ValueSchemaSpec` supports `string`, `number`, `integer`, `boolean`, `null`, `array`, `object`, author-only `json`, and exact-one `oneOf`; scalar `enum` and `const` values must match their node type. An explicit object node always declares `additionalProperties: true | false`. Parameter definitions remain an implicit open object property map, with `required: true` attached to each required property.

Source: [`packages/core/tools/src/schema.ts`](../../packages/core/tools/src/schema.ts)

```ts type-equiv
/** One author-facing schema for any lossless JSON value root. */
type ValueSchemaSpec =
  | StringValueSchemaSpec
  | NumberValueSchemaSpec
  | IntegerValueSchemaSpec
  | BooleanValueSchemaSpec
  | NullValueSchemaSpec
  | ArrayValueSchemaSpec
  | ObjectValueSchemaSpec
  | JsonValueSchemaSpec
  | OneOfValueSchemaSpec
```

```ts type-equiv
/** One implicit parameter-root property, optionally required. */
type ParameterPropertySpec = ValueSchemaSpec & { required?: true }
```

```ts type-equiv
/**
 * Tool parameter schema. The map itself is an implicit open object root;
 * requiredness remains a per-property `required: true` annotation.
 */
type ParameterSchemaSpec = Record<string, ParameterPropertySpec>
```

`{ type: 'json' }` infers `JsonValue` and compiles to an annotation-only unconstrained raw schema. Output roots can be objects, arrays, scalars, or null. `InferValue<S>` honors literal constraints and object openness; `InferArgs<P>` turns per-property requiredness into required and optional keys:

```ts type-equiv
/**
 * Infer the TypeScript value accepted by an author-facing value schema.
 * Output schemas may therefore infer object, array, scalar, or null roots.
 */
type InferValue<S extends ValueSchemaSpec, D extends readonly unknown[] = readonly []> =
  D['length'] extends 12 ? JsonValue :
    S extends StringValueSchemaSpec ? InferScalar<S, string> :
      S extends NumberValueSchemaSpec | IntegerValueSchemaSpec ? InferScalar<S, number> :
        S extends BooleanValueSchemaSpec ? InferScalar<S, boolean> :
          S extends NullValueSchemaSpec ? null :
            S extends ArrayValueSchemaSpec
              ? S extends { items: infer I extends ValueSchemaSpec } ? InferValue<I, NextDepth<D>>[] : JsonValue[]
              : S extends ObjectValueSchemaSpec ? InferObject<S, NextDepth<D>> :
                S extends JsonValueSchemaSpec ? JsonValue :
                  S extends OneOfValueSchemaSpec ? InferValue<S['oneOf'][number], NextDepth<D>> :
                    never
```

```ts type-equiv
/** Infer the TypeScript argument object for an implicit parameter schema. */
type InferArgs<S extends ParameterSchemaSpec> = InferProperties<S, readonly []>
```

`defineTool({ name, description, parameters, output, execute, … })` ties parameter inference to `parameterSchemaSpecToJsonSchema()` and `validateArgs()`, and ties `execute`/`render`/`presentationMeta` to `InferValue<OutputSchema>`. Inference widens to `JsonValue` after twelve nested nodes so large schemas remain compilable; runtime validation keeps walking the complete schema. `valueSchemaSpecToJsonSchema()` compiles output declarations through the same enforced raw subset. A parameter mismatch throws `ToolArgsError` (`INVALID_ARGS`); an invalid body or post-policy value throws `ToolOutputError` (`INVALID_TOOL_OUTPUT`). Both use the normal tool-error path. Raw JSON Schema remains open by default; unsupported keywords reject instead of being accepted without enforcement.

Registration is a trusted same-process contract. The registry borrows the typed definition as readonly input, requires `output`, validates its raw schema, and checks semantic requirements such as a positive finite `timeoutMs`; `schemas()` materializes the explicit model-facing projection at the model boundary so execution and presentation share one resolved definition without leaking callbacks onto the wire.

## `ToolRestriction` — one scope's live global filter

`ToolRestriction` applies only to the live deployment-global tool layer. The registry compiles readonly names into private sets, intersects multiple restrictions, then overlays scope-local tools. A deny-only filter admits later unlisted globals, while an allow-list excludes them.

```ts type-equiv
/**
 * Per-scope filter over global tools. Restrictions intersect and do not affect
 * scoped registrations or the reserved Code Mode transport.
 */
interface ToolRestriction {
  /** Global tool names that stay visible; everything else is removed. */
  readonly allow?: readonly string[]
  /** Global tool names removed from visibility. */
  readonly deny?: readonly string[]
}
```

## Execution: extensible waterfalls plus monotonic policy

`ctx.tools.execute()` accepts a caller-owned `ToolExecutionInput`, materializes its parsed JSON arguments once into a pipeline-owned `ToolExecution`, and runs that call through `tools/pre-execute` (the reorderable allow/deny/ask waterfall) → registered monotonic guards → `tools/execute` (around-dispatch wrappers) → `tools/post-execute` (inspect/replace the result) → `tools/result` (the immutable authoritative outcome). The outcome is a `ToolExecutionResult`.

```ts type-equiv
/** Opaque call identity that permits correlation without exposing mutable execution state. */
type ToolExecutionToken = symbol & { readonly [toolExecutionTokenBrand]: true }
```

```ts type-equiv
/**
 * Caller-supplied description of one tool call. {@link ToolRegistry.execute}
 * adds the registry-owned token to form a pipeline {@link ToolExecution};
 * callers do not choose that token.
 */
interface ToolExecutionInput {
  readonly callId: CallId
  readonly name: string
  /** Losslessly JSON-serializable parsed arguments (tools validate their own schema). */
  readonly arguments: unknown
  /** The agent on whose behalf the call runs (set by the agent loop). */
  readonly agent?: Agent
  /**
   * Opaque token of the enclosing transport execution, when one exists. Code
   * Mode sets this on SDK sub-dispatches so commit-style observers can wait for
   * the outer `run_code` outcome without receiving its live mutable execution.
   */
  readonly parent?: ToolExecutionToken
  signal?: AbortSignal
}
```

A tool body receives the runtime extension. `deferContext()` is the composite-tool channel: it records nested-dispatch context without injecting inside the still-open outer call.

```ts type-equiv
/**
 * Runtime context handed to a tool implementation after the registry has
 * accepted a {@link ToolExecution}. A composite tool uses
 * {@link deferContext} to ferry context produced by nested dispatches back to
 * the outer result; the loop appends it only after the outer `tool/result`.
 */
interface ToolRunContext extends ToolExecution {
  /**
   * Defer one nested-dispatch context until this tool's final result reaches
   * the agent loop. Contexts retain their individual source and metadata and
   * are emitted in call order.
   */
  deferContext(context: HookContext): void
}
```

The agent loop asks the registry for each pending call's execution mode and uses it to form exclusive barriers and rolling-pool parallel runs:

```ts type-equiv
/**
 * Scheduling mode for one pending call. `parallel` may overlap with siblings;
 * `exclusive` runs alone and forms an ordering barrier.
 */
type ToolExecutionMode =
  | { kind: 'parallel' }
  | { kind: 'exclusive' }
```

```ts type-equiv
/**
 * One pending tool call inside the registry pipeline. Parsed arguments cross
 * one lossless-JSON materialization boundary before policy and are deep-frozen;
 * call identity and the registry-assigned {@link token} are readonly. An
 * around-dispatch wrapper may set, replace, or remove `signal`. The registry
 * freezes the complete object before `tools/result` observers run.
 */
interface ToolExecution extends ToolExecutionInput {
  /** Registry-assigned identity shared with nested calls only as their opaque `parent` token. */
  readonly token: ToolExecutionToken
}
```

`ToolExecutionToken` is an opaque runtime `Symbol` used only for identity comparison. Before policy, `execute()` materializes and freezes arguments, rejects non-JSON input, and assigns the token. Identity fields and the optional parent token remain readonly; only `signal` may change around dispatch. Final observers receive the frozen execution identity.

A `ToolGuard` is scope-aware final pre-dispatch policy. Its shape deliberately has no allow result: `undefined` preserves the waterfall decision, while a returned reason can only reduce permission, so a later listener cannot undo it.

```ts type-equiv
/**
 * A monotonic execution guard evaluated after every `tools/pre-execute`
 * listener and before the tool body. Returning a reason denies the call;
 * returning `undefined` leaves it unchanged. Because guards have no allow
 * result, listener ordering cannot turn a denial back into permission.
 * @param execution - the identity-protected call after extensible pre-execute policy completed.
 * @returns a final denial reason, or `undefined` to leave the call allowed.
 */
type ToolGuard = (execution: Readonly<ToolExecution>) => string | undefined
```

```ts type-equiv
/** Canonical failure detail; internal routing information remains optional. */
interface ToolFailure {
  /** Human-readable failure message without the Native `Error: ` envelope. */
  message: string
  /** Internal error class/code used by policy and durable diagnostics. */
  info?: ToolErrorInfo
}
```

```ts type-equiv
/** Successful canonical tool execution, including its Native/model projection. */
interface ToolExecutionSuccess {
  readonly isError: false
  /** Execution-local canonical value; deliberately omitted from durable events. */
  readonly value: JsonValue
  readonly content: ContentBlock[]
  readonly error?: never
  readonly meta?: JsonValue
  readonly additionalContexts?: HookContext[]
}
```

```ts type-equiv
/** Failed canonical tool execution; failures never carry a successful value. */
interface ToolExecutionFailure {
  readonly isError: true
  readonly error: ToolFailure
  readonly value?: never
  readonly content: ContentBlock[]
  readonly meta?: JsonValue
  readonly additionalContexts?: HookContext[]
}
```

```ts type-equiv
/** The discriminated, execution-local outcome of one tool call. */
type ToolExecutionResult = ToolExecutionSuccess | ToolExecutionFailure
```

The result carries only the outcome. Call identity remains on the immutable `ToolExecution` that accompanies it through every hook and on the durable `tool/call` / `tool/result` session events, so wrappers cannot create a second, disagreeing identity. The canonical `value` is execution-local: the loop persists only `content`, `error`, and `meta`, while `tool/code-dispatch` stores a bounded summary. Replay reproduces presentation but cannot reconstruct intermediate values.

On success the registry snapshots and validates the body value, freezes it, and invokes the pure renderer plus the optional direct-surface metadata projector. It separately materializes the durable presentation fields immediately before `tools/result`; an invalid value, renderer/projector failure, or non-JSON presentation becomes a JSON-safe `isError`. The final live observer therefore sees the exact execution-local value beside fields safe for the later durable append.

Each interception waterfall returns a typed **Decision** (the idiom shared with the `agent/*` seams). `tools/pre-execute` listeners receive `(exec, next)` and return a `PreToolDecision`; `tools/execute` wrappers return a `ToolExecutionResult`; `tools/post-execute` listeners receive `(exec, result, next)` and return a `PostToolDecision`:

```ts type-equiv
/**
 * Pre-dispatch decision. `allow` runs the call; `deny` materializes an error;
 * `ask` runs only after an approval service returns `allowed-once` and otherwise
 * denies. Input rewriting is excluded because arguments are already logged and
 * presented.
 */
type PreToolDecision =
  | { kind: 'allow' }
  | { kind: 'deny'; reason: string }
  | { kind: 'ask'; reason?: string }
```

```ts type-equiv
/**
 * Post-dispatch decision: accept, replace one projection, attach context for the
 * next request, or block by turning corrective feedback into an error result.
 */
type PostToolDecision =
  | { kind: 'accept'; content?: ContentBlock[]; value?: never; additionalContexts?: HookContext[] }
  | { kind: 'accept'; value: JsonValue; content?: never; additionalContexts?: HookContext[] }
  | { kind: 'block'; feedback: ContentBlock[]; additionalContexts?: HookContext[] }
```

Call `next()` for the default or return a decision to short-circuit. Pre-policy may deny or ask; only `allowed-once` proceeds, while a non-grant, missing approval channel or service, or agent-less request becomes a denial. Guards may still impose a final denial. Arguments cannot be rewritten because history, audit, UI, and execution must agree.

Post-policy may replace either content or value, never both. Content replacement preserves the canonical value and existing metadata; value replacement is revalidated and recomputes content/metadata; a block removes the value and becomes an `isError` containing corrective feedback. Content replacement is presentation policy, not confidentiality policy: a listener that must hide the programmatic value blocks or replaces it. `tools/result` receives the frozen execution and result after normalization; observers cannot transform them, and observer failures are contained. Unknown and throwing tools both become structured errors (`ToolNotFoundError` maps to `UNKNOWN_TOOL`), so the call fails without ending the turn.

## The enforced raw JSON Schema subset

Raw schemas from subagents, workflows, MCP, and dynamic registrations use the wire-level counterpart of the author DSL. `assertSupportedJsonSchema()` accepts any JSON root, `validateJsonSchemaValue()` enforces it, and `JsonSchemaError` reports every unsupported or malformed schema path. The empty annotation-only node means unconstrained lossless JSON. `oneOf` requires at least two branches and a value must match exactly one. Consumers that still require an object root call `assertObjectJsonSchema()` and carry `ObjectJsonSchema`; this is how subagent/workflow caller-defined structured output remains object-rooted without restricting the shared vocabulary.

```ts type-equiv
/** Scalar JSON values supported by `enum` and `const`. */
type JsonSchemaScalar = string | number | boolean | null
```

```ts type-equiv
/** Single-type keywords accepted by the enforced subset. */
type JsonSchemaType = 'object' | 'array' | 'string' | 'number' | 'integer' | 'boolean' | 'null'
```

```ts type-equiv
/**
 * One raw JSON Schema node in the enforced subset. The optional fields express
 * the external wire shape; {@link assertSupportedJsonSchema} rejects invalid
 * combinations before a caller treats the node as trusted.
 */
interface JsonSchemaNode {
  /** Omit with no constraints for any JSON value, or use `oneOf`. */
  type?: JsonSchemaType
  /** Exactly one branch must validate; at least two branches are required. */
  oneOf?: JsonSchemaNode[]
  /** Nested property schemas (`type: 'object'` only). */
  properties?: Record<string, JsonSchemaNode>
  /** Required property names; each must appear in `properties`. */
  required?: string[]
  /** `false` rejects undeclared keys; absent/`true` follows JSON Schema's open default. */
  additionalProperties?: boolean
  /** Item schema (`type: 'array'` only); absent accepts any JSON item. */
  items?: JsonSchemaNode
  /** Allowed values for a scalar node. */
  enum?: JsonSchemaScalar[]
  /** The single allowed value for a scalar node. */
  const?: JsonSchemaScalar
  /** Annotation, ignored for validation. */
  description?: string
  /** Annotation, ignored for validation. */
  title?: string
  /** Annotation, ignored for validation but required to be lossless JSON. */
  default?: JsonValue
  /** Annotation, ignored for validation but required to be lossless JSON. */
  examples?: JsonValue
}
```

```ts type-equiv
/** A consumer-constrained object-rooted schema. */
type ObjectJsonSchema = JsonSchemaNode & { type: 'object' }
```

## Tool-presentation UI vocabulary

How a tool wants its call shown in a UI (an editor tool-call card, a CLI log line), provider-neutral so a tool describes itself without depending on any client protocol. `presentCall`/`presentResult` return a **`card`-tagged render intent** — a discriminated union a UI bridge switches on:

- `ToolCallView` (pending): `{ card: 'generic', title, kind?, rawInput?, content?, locations? }` (the default card; `locations` is `{ path, line? }[]` files the call reads/modifies, for editor follow-along), `{ card: 'terminal', title, description?, cwd? }` (a shell command → a terminal card), or `{ card: 'diff', title, diffs, locations? }` (a file create/modify → an inline diff card; `diffs` is `{ path, oldText, newText }[]`, `oldText: null` for a new file).
- `ToolResultView` (completed): `{ card: 'generic', title?, content? }`, `{ card: 'terminal', title?, output?, exitCode?, signal? }` (the captured run output + exit; a capable UI shows an exit-status pill, an incapable one gets a fenced ` ```console ` fallback the bridge derives from `output`), or `{ card: 'diff', title?, diffs }` (a completed file mutation → the change to show, typically the applied hunks with context lines computed from the before/after content, or a whole-file diff when there is no before-image — e.g. a file create. A `tool_call_update`'s content REPLACES the call's content, so a mutation tool returns this even when it duplicates the call-time snippet, to keep the result from clobbering the diff with result text).

`ToolCallKind` (`'read' | 'edit' | 'delete' | 'move' | 'search' | 'execute' | 'fetch' | 'other'`) picks an icon on a generic card. `FileLocation` (`{ path, line? }`) and `FileDiff` (`{ path, oldText, newText }`) are the shared file-card vocabulary. The design is pinned in [the render-intent-union Agent Note](../../.agents/notes/implemented/architecture/2026-07-02-tool-render-intent-union.md); the ACP bridge maps a `diff` card to a `{ type: 'diff' }` content block, a `terminal` card to the `_meta` terminal convention, and relativizes a file card's title against the session cwd.

The full presentation field docs live in [`packages/core/tools/src/presentation.ts`](../../packages/core/tools/src/presentation.ts). The `bash` schema and executor are on [bash.md](bash.md); generic background controls are on [tasks.md](tasks.md).