ZiyaZhang
e8f5c0b51b
Minimal Electron shell over the DSH JSON-RPC runtime — a first-look at what a ChatGPT.app-style host on top of the DeepSeek Harness looks like, with the harness's normally-invisible internals (trace timeline, context surface, subagent tree, compaction, plugin registry, rubrics) brought forward as first-class UI surfaces so plugin authors and researchers can see what the agent is actually doing. Runs against three keyless-to-live profiles (stdio-echo works on master out of the box; daemon-echo / daemon-vibe-echo activate once the daemon-demo lands; stdio-deepseek and daemon-vibe hit the real DeepSeek API when you supply a key). HARNESS_DEV auto-resolves to the in-repo runtime when this shell ships under examples/desktop/, so a fresh clone launches without config; env DSH_DEV_ROOT overrides for custom layouts, and a sibling deepseek-harness-dev/ checkout is the original dev workflow. Cold-clone gate (P0 fixes for first-time-clone usability): - HARNESS_DEV: 3-candidate resolver (env → walk-up in-repo marker → sibling), unit-tested via mock fs so ordering is locked without needing either real layout on disk. - config yml leaves rewritten at assemble time so the sibling-clone paths (../../deepseek-harness-dev/examples/echo-agent/…) become the in-repo paths (../../echo-agent/…) in the released tree — source yml stays usable for local dev, released tree ships a working shape. - pnpm-workspace.yaml allowBuilds.electron = true (was placeholder). - missing-key card in stdio-deepseek offers a one-click switch to stdio-echo (the keyless profile that works on master) rather than daemon-echo (blocked on the not-yet-shipped daemon-demo). - assemble-oss-release.sh rewrites the source-side breadcrumb name 'dsh-desktop-demo' → 'dsh-desktop' for the released package.json. FOUC guard on the onboarding gate (41fc5df carried) keeps the first-launch splash from flashing before the runtime probe finishes. Test suite (1634 tests in source, 3990 in the runtime repo) covers resolver ordering, renderer classifiers, trace timeline shape, compaction diff rendering, rubric parity, and the missing-key onboarding paths.
122 lines
6.1 KiB
JSON
122 lines
6.1 KiB
JSON
[
|
|
{
|
|
"_mock": true,
|
|
"_mockReason": "Clickability audit P2 (2026-07-17): reasoning-heavy turn used by mock-reasoning-only to demo the DeepSeek differentiator — reasoning-delta chunks + usage.reasoningTokens are the DSH signal upstream Claude/OpenAI hosts don't surface. One turn, one step, one assistant message; the reasoning body dwarfs the visible reply, and usage carries reasoningTokens=982 so the tab chip reads honestly."
|
|
},
|
|
{
|
|
"type": "user/message",
|
|
"time": 1721501000050,
|
|
"seq": 5000,
|
|
"data": {
|
|
"content": [
|
|
{ "type": "text", "text": "Given the tri-view detail pane, where is the Reasoning tab supposed to live and why can most upstream hosts not show one?" }
|
|
]
|
|
}
|
|
},
|
|
{
|
|
"type": "turn/start",
|
|
"time": 1721501000080,
|
|
"seq": 5001,
|
|
"data": { "turn": 1, "trigger": "user" }
|
|
},
|
|
{
|
|
"type": "step/start",
|
|
"time": 1721501000100,
|
|
"seq": 5002,
|
|
"data": { "turn": 1, "step": 0 }
|
|
},
|
|
{
|
|
"type": "request/header",
|
|
"time": 1721501000110,
|
|
"seq": 5003,
|
|
"data": {
|
|
"header": {
|
|
"model": "deepseek-reasoner",
|
|
"provider": "deepseek",
|
|
"system": "You are a careful research assistant. Think step by step before you answer.",
|
|
"config": { "temperature": 0.3, "maxTokens": 4096 }
|
|
},
|
|
"reason": "step-start"
|
|
}
|
|
},
|
|
{
|
|
"type": "assistant/chunk",
|
|
"time": 1721501000300,
|
|
"seq": 5004,
|
|
"data": { "chunk": { "type": "reasoning-delta", "text": "Let me trace this carefully. The tri-view is Tree / Timeline / Graph, " } }
|
|
},
|
|
{
|
|
"type": "assistant/chunk",
|
|
"time": 1721501000350,
|
|
"seq": 5005,
|
|
"data": { "chunk": { "type": "reasoning-delta", "text": "and each of them lands a right-side detail pane on click. So the Reasoning tab lives inside that right-side pane, alongside Feedback / Input / Output / Attributes. " } }
|
|
},
|
|
{
|
|
"type": "assistant/chunk",
|
|
"time": 1721501000420,
|
|
"seq": 5006,
|
|
"data": { "chunk": { "type": "reasoning-delta", "text": "It's conditional — a step without reasoning-delta chunks or a positive usage.reasoningTokens hides the tab entirely, so casual turns don't get a fifth clickable that says nothing. " } }
|
|
},
|
|
{
|
|
"type": "assistant/chunk",
|
|
"time": 1721501000480,
|
|
"seq": 5007,
|
|
"data": { "chunk": { "type": "reasoning-delta", "text": "As for why upstreams don't show it: Anthropic's Claude ships an `extended_thinking` field on the API but hides the text in the product; OpenAI's o-series does the same with its `reasoning.summary`. " } }
|
|
},
|
|
{
|
|
"type": "assistant/chunk",
|
|
"time": 1721501000550,
|
|
"seq": 5008,
|
|
"data": { "chunk": { "type": "reasoning-delta", "text": "DeepSeek exposes reasoning as a first-class content-block type plus a dedicated reasoning-delta chunk, and DSH's aggregator forwards both to the UI. " } }
|
|
},
|
|
{
|
|
"type": "assistant/chunk",
|
|
"time": 1721501000620,
|
|
"seq": 5009,
|
|
"data": { "chunk": { "type": "reasoning-delta", "text": "So the visualisation happens because the wire carries the payload — the product is honest about what the model produced instead of collapsing it into a summary. That's the differentiator the audit calls out." } }
|
|
},
|
|
{
|
|
"type": "assistant/chunk",
|
|
"time": 1721501000700,
|
|
"seq": 5010,
|
|
"data": { "chunk": { "type": "text-delta", "text": "The Reasoning tab lives in the tri-view's right-side detail pane, next to Feedback / Input / Output / Attributes, and appears only when the step has reasoning-delta content or a positive usage.reasoningTokens. " } }
|
|
},
|
|
{
|
|
"type": "assistant/chunk",
|
|
"time": 1721501000740,
|
|
"seq": 5011,
|
|
"data": { "chunk": { "type": "text-delta", "text": "Most upstream hosts don't expose one because their wire hides reasoning: Anthropic keeps `extended_thinking` on the API only, OpenAI ships an aggregate `reasoning.summary`, and neither routes the raw stream to the UI. DeepSeek does — reasoning is a first-class content-block and its own chunk type — so DSH can render it faithfully." }
|
|
}
|
|
},
|
|
{
|
|
"type": "assistant/message",
|
|
"time": 1721501000800,
|
|
"seq": 5012,
|
|
"data": {
|
|
"content": [
|
|
{ "type": "reasoning", "text": "Let me trace this carefully. The tri-view is Tree / Timeline / Graph, and each of them lands a right-side detail pane on click. So the Reasoning tab lives inside that right-side pane, alongside Feedback / Input / Output / Attributes. It's conditional — a step without reasoning-delta chunks or a positive usage.reasoningTokens hides the tab entirely, so casual turns don't get a fifth clickable that says nothing. As for why upstreams don't show it: Anthropic's Claude ships an `extended_thinking` field on the API but hides the text in the product; OpenAI's o-series does the same with its `reasoning.summary`. DeepSeek exposes reasoning as a first-class content-block type plus a dedicated reasoning-delta chunk, and DSH's aggregator forwards both to the UI. So the visualisation happens because the wire carries the payload — the product is honest about what the model produced instead of collapsing it into a summary. That's the differentiator the audit calls out." },
|
|
{ "type": "text", "text": "The Reasoning tab lives in the tri-view's right-side detail pane, next to Feedback / Input / Output / Attributes, and appears only when the step has reasoning-delta content or a positive usage.reasoningTokens. Most upstream hosts don't expose one because their wire hides reasoning: Anthropic keeps `extended_thinking` on the API only, OpenAI ships an aggregate `reasoning.summary`, and neither routes the raw stream to the UI. DeepSeek does — reasoning is a first-class content-block and its own chunk type — so DSH can render it faithfully." }
|
|
],
|
|
"usage": {
|
|
"inputTokens": 214,
|
|
"outputTokens": 118,
|
|
"reasoningTokens": 982,
|
|
"cacheReadTokens": 0
|
|
},
|
|
"finish_reason": "stop"
|
|
}
|
|
},
|
|
{
|
|
"type": "step/end",
|
|
"time": 1721501000900,
|
|
"seq": 5013,
|
|
"data": { "turn": 1, "step": 0, "reason": { "kind": "stop" } }
|
|
},
|
|
{
|
|
"type": "turn/end",
|
|
"time": 1721501000950,
|
|
"seq": 5014,
|
|
"data": { "turn": 1, "reason": { "kind": "stop" } }
|
|
}
|
|
]
|