Caller-side bridge from agent frameworks to the official DeepSeek Harness Web Host.
Server defines 9 tools with explicit schemas, descriptions, and type annotations. Tool names follow verb_noun pattern (dsh_delegate, dsh_followup, dsh_status). All tools have descriptions (50-200 chars, within baseline). Input schemas use JSON Schema with types, enums, and regex patterns. However, output schemas are NOT documented, responses are JSON-stringified but structure is opaque to the LLM. Error handling is present (DshRpcError, DshTransportError, etc.) but recovery guidance is minimal. No parameter descriptions explain when to use dsh_followup vs dsh_continue (both are aliases). Tool composition is reasonable but lacks chaining hints (e.g., dsh_delegate returns taskId, which dsh_status accepts, this dependency should be explicit in descriptions).
Respond to a pending question request
Compatibility alias for dsh_followup. Continue the same root DSH session. queue targets the next turn; steer targets the active turn's next step. The write is never automatically retried.
Create a root session on the configured official DSH Web Host and queue the initial prompt. Uses DSH's configured model; no model argument is accepted. Detached by default. workspaceMode is only a bridge-local cooperative claim and does not select or verify the DSH sandbox.
Continue the same root DSH session. queue targets the next turn; steer targets the active turn's next step. The write is never automatically retried.
Report the connect-only bridge state and current official DSH Web Host capabilities.
Respond to a pending approval request
Output schemas not documented. Tools return JSON-stringified responses (result() function) but LLM cannot infer field structure, types, or chaining IDs. Agents cannot plan downstream calls without guessing response shape.
dsh_continue is a duplicate alias of dsh_followup with identical schema and nearly identical description. LLM cannot distinguish when to use each; wastes reasoning cycles. Violates single-responsibility principle.
dsh_tail description is vague ('Read bounded event digests using bridge coordination'). Does not explain what 'event digests' are, when to call it vs dsh_wait, or what structure is returned. Insufficient for LLM selection.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 61 | 2026-07-28+ | v2 |
Return separate availability/execution state, root and descendant sessions, queue depths, pending interactions, final message, bridge cursor/watermarks, and bridge-local workspace claim semantics.
Read bounded event digests using bridge coordination
Wait for task state changes with configurable timeout and wake conditions
Error handling returns structured error objects (DshRpcError, DshTransportError) but does not guide recovery. No 'try this next' hints. LLM receives error code but cannot infer whether to retry, ask user, or abort.
Tool descriptions lack dependency hints. dsh_delegate returns taskId; dsh_status accepts taskId. This chaining relationship is not documented, forcing LLM to infer it.