The first real LlmAdapter implementations, shipped as a deliberate pair:
same models and wire protocol, completely different internals, so the
StreamChunk protocol is verified across independent implementations.
- dsh-llm-deepseek: hand-rolled fetch + SSE parser + chunk-translation
state machine against the official chat-completions format (thinking
mode via top-level thinking/reasoning_effort; the empty-string
reasoning_content first chunk; usage attached to the finish chunk or
trailing; reasoning_content passback on tool-call turns; disjoint
cache-token accounting).
- dsh-llm-pi-ai: the same endpoint through @earendil-works/pi-ai,
mapping its event vocabulary (parsed tool arguments, in-stream error
events, folded reasoning tokens) onto the same chunks.
The agent loop now honors the in-band error path: an adapter that ends
its stream with finish {kind:error|aborted} (the only option for
adapters that can't throw mid-stream, like pi-ai) is translated into a
step error, so the turn ends error/aborted with a logged error event
instead of a normal completed assistant message. This makes the
StreamChunk error contract real for both adapters; docs/architecture.md
and the StreamChunk doc are updated accordingly.
New yarn test:e2e (vitest.e2e.config.ts, *.e2e.ts) runs key-gated
real-API matrices for both adapters across V4 Flash/Pro and all
thinking/effort levels; it self-skips without DEEPSEEK_API_KEY. Unit
suites run against local node:http mock SSE servers at 100% per-file
coverage.
Three packages following the new capability-seam pattern (interface /
implementation / consumer, now documented in docs/architecture.md):
- dsh-bash: abstract BashExecutor service (ctx.bash) + vocabulary types.
- dsh-bash-local: local subprocesses — bash -c per call in a detached
process group, SIGTERM→SIGKILL group kills, tail-keep truncation with
full-stream spill files, model-friendly env, background task registry.
- dsh-tool-bash: the bash / bash_output / bash_kill tool schemas with
runtime arg validation and background completion notices via
agent.inject(). Non-zero exits are reported, not errored.
Design surveyed against the bash tools of Claude Code, OpenCode, Codex,
and pi (notes in the package READMEs). Permissions/sandbox stay TODO on
the tools/execute waterfall seam; stateful-shell alternatives recorded
in run.ts.
InferArgs now produces genuinely optional keys: required/optional
properties are split at the key level (RequiredKeys + mapped `?`), so
{ limit: { type: 'number' } } infers as { limit?: number } and callers
can omit it — previously the key stayed required with `| undefined`.
Array item inference recurses (arrays of objects infer their element
shape instead of Record<string, unknown>), matching the generated
JSON Schema.
Tool execution error reporting handles non-Error throws again:
`throw { message: 'denied' }` reports the message instead of
"[object Object]" (errorMessage helper).
The new schema tests now actually typecheck: schema literals use
`satisfies SchemaSpec` (the standalone-literal widening made
schemaSpecToJsonSchema reject the suite's own examples), and the
ToolSchema probe cast goes through unknown. Tests-and-examples
typechecking is now part of `yarn typecheck` via the new
tsconfig.typecheck.json (resolves vendor packages by their built
declarations, so vendor's relaxed-strictness source stays out of
scope) — vitest never typechecks, so this gate is what catches such
breakage. +4 regression tests (typed omission, array-of-objects
inference both type- and runtime-level, non-Error throw message).
Tianyi Cui