Add web multimodal image attachments

docs/core-data-structures/attachment.md

@@ -0,0 +1,70 @@
+# Durable Image Attachments
+
+The attachment seam separates binary image ownership from the session log. A producer gives validated encoded bytes to [`ctx.attachments`](../cordis-catalog/services.md#ctxattachments); the service publishes an immutable content-addressed reference only after the object is durable. Session events and model-visible `ImageBlock`s contain that reference and metadata, never a browser object URL, host temporary path, provider URL, or base64 payload.
+
+Unsent browser drafts may stay in memory and native clients may stage them in operating-system temporary storage. Once the host accepts a user message, its images move below `<DSH_HOME>/attachments/v1` before the user event is appended. Structured model image output follows the same persist-before-event rule.
+
+Source: [`packages/attachment/attachment/src/types.ts`](../../packages/attachment/attachment/src/types.ts)
+
+## Identity and verified metadata
+
+`AttachmentId` is a branded opaque string. The local backend currently emits `sha256:<digest>`, but consumers must neither parse that representation nor derive a filesystem path from it.
+
+```ts type-equiv
+/** Raster image formats accepted by the version-one attachment path. */
+type ImageMediaType = 'image/png' | 'image/jpeg' | 'image/webp' | 'image/gif'
+```
+
+```ts type-equiv
+/** Durable, serializable metadata for one immutable image object. */
+interface ImageAttachmentRef {
+  /** Opaque storage identifier; never a filesystem path or bearer URL. */
+  attachmentId: AttachmentId
+  /** Media type verified from the stored bytes. */
+  mediaType: ImageMediaType
+  /** Exact encoded byte length. */
+  bytes: number
+  /** Intrinsic encoded width in pixels. */
+  width: number
+  /** Intrinsic encoded height in pixels. */
+  height: number
+  /** Optional display name stripped of local path information. */
+  name?: string
+}
+```
+
+```ts type-equiv
+/** Deployment-resolved limits shared by upload consumers and UI preflight. */
+interface ImageAttachmentLimits {
+  maxImageBytes: number
+  maxImagesPerMessage: number
+  maxMessageImageBytes: number
+  maxImagePixels: number
+  mediaTypes: readonly ImageMediaType[]
+}
+```
+
+The reference records intrinsic dimensions and encoded length so clients can lay out history without decoding first, while every authoritative read still re-checks digest, media signature, dimensions, and metadata against the object.
+
+## Commit and verified-read payloads
+
+```ts type-equiv
+/** Request to validate and durably commit one image. */
+interface SaveImageAttachment {
+  data: Uint8Array
+  /** Caller-declared media type, checked against magic bytes. */
+  mediaType: ImageMediaType
+  /** Optional browser/provider display name; it is never interpreted as a path. */
+  name?: string
+}
+```
+
+```ts type-equiv
+/** Stored image bytes returned after reference and digest verification. */
+interface StoredImageAttachment {
+  ref: ImageAttachmentRef
+  data: Uint8Array
+}
+```
+
+`saveImage()` validates bytes and atomically commits one object before returning its reference. `readImage()` accepts a reference from an authorized session path and returns bytes only after integrity verification. The service is deliberately retention-neutral: resumed and forked sessions may share objects, so reference-aware garbage collection is deferred rather than tied to any one session's deletion.

docs/core-data-structures/core.md

@@ -28,6 +28,7 @@ Everything else is documented on a **sub-page**, not here. The rule that draws t
 | [tools.md](tools.md) | `ToolDefinition` full fields, the schema DSL, `ToolExecution`/`ToolResult`, tool-presentation UI types, and the guarded execution pipeline |
 | [user-interaction.md](user-interaction.md) | the UI-backed human question/answer seam: `AskUserQuestionRequest`, answer/options vocabulary, provider API, error taxonomy |
 | [approval.md](approval.md) | the one-shot user-approval seam: `ApprovalRequest`, `ApprovalOutcome`, per-session policy, audit and answerer contracts |
+| [attachment.md](attachment.md) | durable image identity and metadata, validation inputs, verified reads, and the `AttachmentStore` seam |
 | [bash.md](bash.md) | the bash executor seam: `BashExecRequest`/`Spec`, `BashRunResult`, background `BashProcess` handles |
 | [pty.md](pty.md) | persistent terminal ids, backend/session contracts, send readiness, bounded reads, and owner-visible snapshots |
 | [sandbox.md](sandbox.md) | per-session policy resolution and the process-confinement seam: file-effect modes, execution/provider policies, `ConfinedArgv`, enforcement and fail-closed errors |
@@ -106,12 +107,13 @@ Source: [`packages/llm/llm/src/types.ts`](../../packages/llm/llm/src/types.ts)
 interface ContentBlockMap {
   'text': TextBlock
   'reasoning': ReasoningBlock
+  'image': ImageBlock
   'tool-call': ToolCallBlock
   'tool-result': ToolResultBlock
 }
 ```
 
-The block interfaces (full fields in source): `TextBlock` (`text`), `ReasoningBlock` (thinking, distinct from visible text), `ToolCallBlock` (`id: CallId`, `name`, raw-JSON `arguments`), `ToolResultBlock` (`toolCallId`, nested `content: ContentBlock[]`, `isError?`). `ContentBlock = ContentBlockMap[ContentBlockType]`. The core set is limited to blocks every shipping path honors — multimodal content (images, audio, …) has no core block type; a feature that needs one adds it via the merge-extensible map together with the adapter/UI/compaction support that honors it.
+The block interfaces (full fields in source): `TextBlock` (`text`), `ReasoningBlock` (thinking, distinct from visible text), `ImageBlock` (a durable [image attachment](attachment.md) plus optional alternative text), `ToolCallBlock` (`id: CallId`, `name`, raw-JSON `arguments`), and `ToolResultBlock` (`toolCallId`, nested `content: ContentBlock[]`, `isError?`). `ContentBlock = ContentBlockMap[ContentBlockType]`. A new modality belongs in the merge-extensible map only when its adapter, UI, compaction, and durable replay paths honor it.
 
 A `Message` is a role plus blocks. Loop-derived assistant messages carry their durable provider/model identity and optional adapter-private replay metadata:
 
@@ -192,6 +194,10 @@ interface LlmModelInfo {
   name: string
   /** Optional user-facing distinction from otherwise similar models. */
   description?: string
+  /** Accepted request modalities; absent means unknown, while an explicit omission is negative capability. */
+  inputModalities?: readonly ModelModality[]
+  /** Structured response modalities; absent means unknown, while an explicit omission is negative capability. */
+  outputModalities?: readonly ModelModality[]
 }
 ```
 

docs/core-data-structures/llm-streaming.md

@@ -208,6 +208,7 @@ declare abstract class LlmAdapter {
 interface ContentBlockMap {
   'text': TextBlock
   'reasoning': ReasoningBlock
+  'image': ImageBlock
   'tool-call': ToolCallBlock
   'tool-result': ToolResultBlock
 }