feat(agent-stream): thinking and tool streaming

.agents/skills/add-model/SKILL.md

@@ -52,6 +52,7 @@ Use a precise WebFetch prompt: *"Extract for {model_id}: exact model id string,
 | `reasoningEffort` | `openai/core.ts`, `azure-openai`, `anthropic/core.ts` (mapped to thinking), `gemini/core.ts` | **Dead on xai, deepseek, mistral, groq, cerebras, openrouter, fireworks, bedrock, vertex** unless their core consumes it — re-grep before assuming |
 | `verbosity` | `openai/core.ts`, `azure-openai/index.ts` only | Dead elsewhere |
 | `thinking` | `anthropic/core.ts`, `gemini/core.ts` | Dead elsewhere |
+| `thinking.streamed` | Docs generator + `getThinkingStreamVisibility` (`models.ts`); `anthropic/core.ts` uses `'summary'` to request `display: 'summarized'` on agent-events runs | **Mandatory on Anthropic-family thinking models** (`agent-stream-docs:check` fails without it); other families fall back to provider defaults |
 | `nativeStructuredOutputs` | `anthropic/core.ts`, `fireworks/index.ts`, `openrouter/index.ts` | Dead on openai, xai, google, vertex, bedrock, azure-openai, deepseek, mistral, groq, cerebras |
 | `maxOutputTokens` | Read by UI + executor for token estimation | Always meaningful — set if provider documents a cap |
 | `computerUse` | `anthropic/core.ts` | Dead elsewhere |
@@ -146,6 +147,15 @@ If anything matches, run the affected provider tests and update assertions as ne
 
 The Consumption Matrix (Step 2) tells you which capability *flags* are honored by existing provider code. But if the new model needs **net-new** request handling that the provider doesn't implement yet — a new beta header (e.g. Anthropic's `anthropic-beta` structured-outputs header in `anthropic/index.ts`), a new thinking/reasoning encoding, a Responses-API quirk — you must edit `apps/sim/providers/<provider>/core.ts` / `index.ts`. Setting a flag whose behavior isn't implemented is a silent no-op. When you do edit provider code, reuse the shared helpers rather than hand-rolling: streaming responses are assembled via `createStreamingExecution` (`@/providers/streaming-execution`) and tool schemas via `adaptOpenAIChatToolSchema` / `adaptAnthropicToolSchema` (`@/providers/tool-schema-adapter`).
 
+### Thinking/reasoning models: `streamed` visibility + generated docs
+
+If the entry has `capabilities.thinking` or `capabilities.reasoningEffort`, it appears in the autogenerated "Streamed thinking and tool calls" table on the Agent block docs page:
+
+- **Anthropic-family (`anthropic`, `azure-anthropic`) thinking models MUST declare `capabilities.thinking.streamed`** (`'full' | 'summary' | 'none'`) — visibility varies per Claude generation. Verify against the provider's thinking/display docs (e.g. Anthropic's "controlling thinking display" page and per-model "what's new" notes): generations that default `thinking.display` to `omitted` (Opus 4.7+, Sonnet 5, Fable 5) are `'summary'` — Sim opts back in with `display: 'summarized'` on agent-events runs; older generations that return full thinking deltas are `'full'`. `bun run agent-stream-docs:check` (CI) fails if the field is missing.
+- Other families usually omit the field and inherit the provider default in `getThinkingStreamVisibility` (Gemini/OpenAI → summaries, Bedrock → none, OpenAI-compat vendors → full raw CoT). Set it explicitly only when the model deviates from its family.
+- After inserting the entry, run `bun run agent-stream-docs:generate` and commit the regenerated `apps/docs/content/docs/en/workflows/blocks/agent.mdx` — CI diffs it.
+- Include the `streamed` value (with its source URL) in the verification report when set.
+
 ### Wrong family entirely?
 
 - **Embedding or rerank model** → it does NOT go in the `models[]` array. Use `EMBEDDING_MODEL_PRICING` / `RERANK_MODEL_PRICING` in `models.ts` instead.
@@ -155,6 +165,7 @@ The Consumption Matrix (Step 2) tells you which capability *flags* are honored b
 
 ```bash
 bun run lint
+bun run agent-stream-docs:generate   # only when the entry has thinking/reasoningEffort
 ```
 
 Lint must pass before reporting done. **If lint fails:** read the error, fix the syntax/typing issue in the entry you just wrote (do not delete the entry — it's the work product), re-run lint, and note the fix in a "Lint adjustments" line in the verification report. Never report done with lint failing.
@@ -201,6 +212,7 @@ Omitting a field is **not the same as verifying it**. Any field you cannot confi
 - ❌ Trusting a marketing email (xAI's grok-4.3 email claimed "3 reasoning efforts" but the API rejects `reasoning_effort` — verified by official docs only)
 - ❌ Setting `nativeStructuredOutputs: true` on xai/openai/google (dead — only anthropic/fireworks/openrouter consume it)
 - ❌ Setting `thinking` on non-Anthropic/non-Gemini providers
+- ❌ Adding an Anthropic-family thinking model without `thinking.streamed` (CI `agent-stream-docs:check` fails), or skipping `bun run agent-stream-docs:generate` after adding any thinking/reasoning model
 - ❌ Setting `verbosity` on anything other than OpenAI gpt-5.x
 - ❌ Copying `pricing.updatedAt` from a sibling instead of using today's date
 - ❌ Inventing a `cachedInput` price by dividing input by 4 (varies by provider — find an explicit number)

.agents/skills/validate-model/SKILL.md

@@ -89,6 +89,7 @@ For each model, evaluate every row. Statuses: ✓ matches docs, ✗ disagrees, 
 - [ ] `reasoningEffort.values` — list matches docs; **omitted** for always-reasoning models that reject the parameter (e.g., grok-4.3, where xAI docs explicitly state `reasoning_effort` is not supported). Verify per model — some always-reasoning models (e.g., OpenAI's o-series) DO accept `reasoning_effort` and should keep the flag.
 - [ ] `verbosity.values` — only on OpenAI gpt-5.x family; values match docs
 - [ ] `thinking.levels` + `thinking.default` — only on Anthropic/Gemini; values match docs
+- [ ] `thinking.streamed` — REQUIRED on Anthropic-family thinking models (`'full'` for generations returning full thinking deltas, `'summary'` for omitted-display generations like Opus 4.7+/Sonnet 5/Fable 5 where Sim requests `display: 'summarized'`); verify against the provider's thinking-display docs. After any change, run `bun run agent-stream-docs:generate` so the Agent block docs table stays in sync (CI diffs it)
 - [ ] `nativeStructuredOutputs` — only on anthropic/fireworks/openrouter; provider must document Structured Outputs / JSON-mode for this model
 - [ ] `toolUsageControl` — provider supports `tool_choice` semantics
 - [ ] `computerUse` — provider implements computer-use loop AND model is a computer-use SKU

.claude/commands/add-model.md

@@ -51,6 +51,7 @@ Use a precise WebFetch prompt: *"Extract for {model_id}: exact model id string,
 | `reasoningEffort` | `openai/core.ts`, `azure-openai`, `anthropic/core.ts` (mapped to thinking), `gemini/core.ts` | **Dead on xai, deepseek, mistral, groq, cerebras, openrouter, fireworks, bedrock, vertex** unless their core consumes it — re-grep before assuming |
 | `verbosity` | `openai/core.ts`, `azure-openai/index.ts` only | Dead elsewhere |
 | `thinking` | `anthropic/core.ts`, `gemini/core.ts` | Dead elsewhere |
+| `thinking.streamed` | Docs generator + `getThinkingStreamVisibility` (`models.ts`); `anthropic/core.ts` uses `'summary'` to request `display: 'summarized'` on agent-events runs | **Mandatory on Anthropic-family thinking models** (`agent-stream-docs:check` fails without it); other families fall back to provider defaults |
 | `nativeStructuredOutputs` | `anthropic/core.ts`, `fireworks/index.ts`, `openrouter/index.ts` | Dead on openai, xai, google, vertex, bedrock, azure-openai, deepseek, mistral, groq, cerebras |
 | `maxOutputTokens` | Read by UI + executor for token estimation | Always meaningful — set if provider documents a cap |
 | `computerUse` | `anthropic/core.ts` | Dead elsewhere |
@@ -145,6 +146,15 @@ If anything matches, run the affected provider tests and update assertions as ne
 
 The Consumption Matrix (Step 2) tells you which capability *flags* are honored by existing provider code. But if the new model needs **net-new** request handling that the provider doesn't implement yet — a new beta header (e.g. Anthropic's `anthropic-beta` structured-outputs header in `anthropic/index.ts`), a new thinking/reasoning encoding, a Responses-API quirk — you must edit `apps/sim/providers/<provider>/core.ts` / `index.ts`. Setting a flag whose behavior isn't implemented is a silent no-op. When you do edit provider code, reuse the shared helpers rather than hand-rolling: streaming responses are assembled via `createStreamingExecution` (`@/providers/streaming-execution`) and tool schemas via `adaptOpenAIChatToolSchema` / `adaptAnthropicToolSchema` (`@/providers/tool-schema-adapter`).
 
+### Thinking/reasoning models: `streamed` visibility + generated docs
+
+If the entry has `capabilities.thinking` or `capabilities.reasoningEffort`, it appears in the autogenerated "Streamed thinking and tool calls" table on the Agent block docs page:
+
+- **Anthropic-family (`anthropic`, `azure-anthropic`) thinking models MUST declare `capabilities.thinking.streamed`** (`'full' | 'summary' | 'none'`) — visibility varies per Claude generation. Verify against the provider's thinking/display docs (e.g. Anthropic's "controlling thinking display" page and per-model "what's new" notes): generations that default `thinking.display` to `omitted` (Opus 4.7+, Sonnet 5, Fable 5) are `'summary'` — Sim opts back in with `display: 'summarized'` on agent-events runs; older generations that return full thinking deltas are `'full'`. `bun run agent-stream-docs:check` (CI) fails if the field is missing.
+- Other families usually omit the field and inherit the provider default in `getThinkingStreamVisibility` (Gemini/OpenAI → summaries, Bedrock → none, OpenAI-compat vendors → full raw CoT). Set it explicitly only when the model deviates from its family.
+- After inserting the entry, run `bun run agent-stream-docs:generate` and commit the regenerated `apps/docs/content/docs/en/workflows/blocks/agent.mdx` — CI diffs it.
+- Include the `streamed` value (with its source URL) in the verification report when set.
+
 ### Wrong family entirely?
 
 - **Embedding or rerank model** → it does NOT go in the `models[]` array. Use `EMBEDDING_MODEL_PRICING` / `RERANK_MODEL_PRICING` in `models.ts` instead.
@@ -154,6 +164,7 @@ The Consumption Matrix (Step 2) tells you which capability *flags* are honored b
 
 ```bash
 bun run lint
+bun run agent-stream-docs:generate   # only when the entry has thinking/reasoningEffort
 ```
 
 Lint must pass before reporting done. **If lint fails:** read the error, fix the syntax/typing issue in the entry you just wrote (do not delete the entry — it's the work product), re-run lint, and note the fix in a "Lint adjustments" line in the verification report. Never report done with lint failing.
@@ -200,6 +211,7 @@ Omitting a field is **not the same as verifying it**. Any field you cannot confi
 - ❌ Trusting a marketing email (xAI's grok-4.3 email claimed "3 reasoning efforts" but the API rejects `reasoning_effort` — verified by official docs only)
 - ❌ Setting `nativeStructuredOutputs: true` on xai/openai/google (dead — only anthropic/fireworks/openrouter consume it)
 - ❌ Setting `thinking` on non-Anthropic/non-Gemini providers
+- ❌ Adding an Anthropic-family thinking model without `thinking.streamed` (CI `agent-stream-docs:check` fails), or skipping `bun run agent-stream-docs:generate` after adding any thinking/reasoning model
 - ❌ Setting `verbosity` on anything other than OpenAI gpt-5.x
 - ❌ Copying `pricing.updatedAt` from a sibling instead of using today's date
 - ❌ Inventing a `cachedInput` price by dividing input by 4 (varies by provider — find an explicit number)

.claude/commands/validate-model.md

@@ -88,6 +88,7 @@ For each model, evaluate every row. Statuses: ✓ matches docs, ✗ disagrees, 
 - [ ] `reasoningEffort.values` — list matches docs; **omitted** for always-reasoning models that reject the parameter (e.g., grok-4.3, where xAI docs explicitly state `reasoning_effort` is not supported). Verify per model — some always-reasoning models (e.g., OpenAI's o-series) DO accept `reasoning_effort` and should keep the flag.
 - [ ] `verbosity.values` — only on OpenAI gpt-5.x family; values match docs
 - [ ] `thinking.levels` + `thinking.default` — only on Anthropic/Gemini; values match docs
+- [ ] `thinking.streamed` — REQUIRED on Anthropic-family thinking models (`'full'` for generations returning full thinking deltas, `'summary'` for omitted-display generations like Opus 4.7+/Sonnet 5/Fable 5 where Sim requests `display: 'summarized'`); verify against the provider's thinking-display docs. After any change, run `bun run agent-stream-docs:generate` so the Agent block docs table stays in sync (CI diffs it)
 - [ ] `nativeStructuredOutputs` — only on anthropic/fireworks/openrouter; provider must document Structured Outputs / JSON-mode for this model
 - [ ] `toolUsageControl` — provider supports `tool_choice` semantics
 - [ ] `computerUse` — provider implements computer-use loop AND model is a computer-use SKU

.cursor/commands/add-model.md

@@ -46,6 +46,7 @@ Use a precise WebFetch prompt: *"Extract for {model_id}: exact model id string,
 | `reasoningEffort` | `openai/core.ts`, `azure-openai`, `anthropic/core.ts` (mapped to thinking), `gemini/core.ts` | **Dead on xai, deepseek, mistral, groq, cerebras, openrouter, fireworks, bedrock, vertex** unless their core consumes it — re-grep before assuming |
 | `verbosity` | `openai/core.ts`, `azure-openai/index.ts` only | Dead elsewhere |
 | `thinking` | `anthropic/core.ts`, `gemini/core.ts` | Dead elsewhere |
+| `thinking.streamed` | Docs generator + `getThinkingStreamVisibility` (`models.ts`); `anthropic/core.ts` uses `'summary'` to request `display: 'summarized'` on agent-events runs | **Mandatory on Anthropic-family thinking models** (`agent-stream-docs:check` fails without it); other families fall back to provider defaults |
 | `nativeStructuredOutputs` | `anthropic/core.ts`, `fireworks/index.ts`, `openrouter/index.ts` | Dead on openai, xai, google, vertex, bedrock, azure-openai, deepseek, mistral, groq, cerebras |
 | `maxOutputTokens` | Read by UI + executor for token estimation | Always meaningful — set if provider documents a cap |
 | `computerUse` | `anthropic/core.ts` | Dead elsewhere |
@@ -140,6 +141,15 @@ If anything matches, run the affected provider tests and update assertions as ne
 
 The Consumption Matrix (Step 2) tells you which capability *flags* are honored by existing provider code. But if the new model needs **net-new** request handling that the provider doesn't implement yet — a new beta header (e.g. Anthropic's `anthropic-beta` structured-outputs header in `anthropic/index.ts`), a new thinking/reasoning encoding, a Responses-API quirk — you must edit `apps/sim/providers/<provider>/core.ts` / `index.ts`. Setting a flag whose behavior isn't implemented is a silent no-op. When you do edit provider code, reuse the shared helpers rather than hand-rolling: streaming responses are assembled via `createStreamingExecution` (`@/providers/streaming-execution`) and tool schemas via `adaptOpenAIChatToolSchema` / `adaptAnthropicToolSchema` (`@/providers/tool-schema-adapter`).
 
+### Thinking/reasoning models: `streamed` visibility + generated docs
+
+If the entry has `capabilities.thinking` or `capabilities.reasoningEffort`, it appears in the autogenerated "Streamed thinking and tool calls" table on the Agent block docs page:
+
+- **Anthropic-family (`anthropic`, `azure-anthropic`) thinking models MUST declare `capabilities.thinking.streamed`** (`'full' | 'summary' | 'none'`) — visibility varies per Claude generation. Verify against the provider's thinking/display docs (e.g. Anthropic's "controlling thinking display" page and per-model "what's new" notes): generations that default `thinking.display` to `omitted` (Opus 4.7+, Sonnet 5, Fable 5) are `'summary'` — Sim opts back in with `display: 'summarized'` on agent-events runs; older generations that return full thinking deltas are `'full'`. `bun run agent-stream-docs:check` (CI) fails if the field is missing.
+- Other families usually omit the field and inherit the provider default in `getThinkingStreamVisibility` (Gemini/OpenAI → summaries, Bedrock → none, OpenAI-compat vendors → full raw CoT). Set it explicitly only when the model deviates from its family.
+- After inserting the entry, run `bun run agent-stream-docs:generate` and commit the regenerated `apps/docs/content/docs/en/workflows/blocks/agent.mdx` — CI diffs it.
+- Include the `streamed` value (with its source URL) in the verification report when set.
+
 ### Wrong family entirely?
 
 - **Embedding or rerank model** → it does NOT go in the `models[]` array. Use `EMBEDDING_MODEL_PRICING` / `RERANK_MODEL_PRICING` in `models.ts` instead.
@@ -149,6 +159,7 @@ The Consumption Matrix (Step 2) tells you which capability *flags* are honored b
 
 ```bash
 bun run lint
+bun run agent-stream-docs:generate   # only when the entry has thinking/reasoningEffort
 ```
 
 Lint must pass before reporting done. **If lint fails:** read the error, fix the syntax/typing issue in the entry you just wrote (do not delete the entry — it's the work product), re-run lint, and note the fix in a "Lint adjustments" line in the verification report. Never report done with lint failing.
@@ -195,6 +206,7 @@ Omitting a field is **not the same as verifying it**. Any field you cannot confi
 - ❌ Trusting a marketing email (xAI's grok-4.3 email claimed "3 reasoning efforts" but the API rejects `reasoning_effort` — verified by official docs only)
 - ❌ Setting `nativeStructuredOutputs: true` on xai/openai/google (dead — only anthropic/fireworks/openrouter consume it)
 - ❌ Setting `thinking` on non-Anthropic/non-Gemini providers
+- ❌ Adding an Anthropic-family thinking model without `thinking.streamed` (CI `agent-stream-docs:check` fails), or skipping `bun run agent-stream-docs:generate` after adding any thinking/reasoning model
 - ❌ Setting `verbosity` on anything other than OpenAI gpt-5.x
 - ❌ Copying `pricing.updatedAt` from a sibling instead of using today's date
 - ❌ Inventing a `cachedInput` price by dividing input by 4 (varies by provider — find an explicit number)

.cursor/commands/validate-model.md

@@ -83,6 +83,7 @@ For each model, evaluate every row. Statuses: ✓ matches docs, ✗ disagrees, 
 - [ ] `reasoningEffort.values` — list matches docs; **omitted** for always-reasoning models that reject the parameter (e.g., grok-4.3, where xAI docs explicitly state `reasoning_effort` is not supported). Verify per model — some always-reasoning models (e.g., OpenAI's o-series) DO accept `reasoning_effort` and should keep the flag.
 - [ ] `verbosity.values` — only on OpenAI gpt-5.x family; values match docs
 - [ ] `thinking.levels` + `thinking.default` — only on Anthropic/Gemini; values match docs
+- [ ] `thinking.streamed` — REQUIRED on Anthropic-family thinking models (`'full'` for generations returning full thinking deltas, `'summary'` for omitted-display generations like Opus 4.7+/Sonnet 5/Fable 5 where Sim requests `display: 'summarized'`); verify against the provider's thinking-display docs. After any change, run `bun run agent-stream-docs:generate` so the Agent block docs table stays in sync (CI diffs it)
 - [ ] `nativeStructuredOutputs` — only on anthropic/fireworks/openrouter; provider must document Structured Outputs / JSON-mode for this model
 - [ ] `toolUsageControl` — provider supports `tool_choice` semantics
 - [ ] `computerUse` — provider implements computer-use loop AND model is a computer-use SKU

.github/workflows/test-build.yml

@@ -140,6 +140,9 @@ jobs:
       - name: Verify skill projections are in sync
         run: bun run skills:check
 
+      - name: Verify agent stream capability docs are in sync
+        run: bun run agent-stream-docs:check
+
       - name: Migration safety (zero-downtime) audit
         run: |
           if [ "${{ github.event_name }}" = "pull_request" ]; then