Inject a file from the managed workspace into the conversation so the model can read its contents directly.

Use this ONLY when the user explicitly asks you to look at, describe, caption, analyse, or otherwise read a file or image that exists in the workspace. Examples of valid use:
- "What does this image show?" / "Describe this photo"
- "Can you read the contents of that PDF?"
- "Summarise the document I uploaded"
- "Caption this screenshot"

Do NOT use this for sending a file to the user — use filesystem action=send for that.
Do NOT use this speculatively or pre-emptively without the user asking you to read/describe a file.

Parameters:
- as=image: enables vision analysis of images (model must support vision). Only valid for image MIME types.
- as=file: includes the file for text/document reading.
- role: which conversation role the injected content appears as (user or system). Defaults to user.
- text: optional text prepended before the injected file or image part.
- mime: optional MIME type hint if auto-detection is unreliable.
- filename: optional display filename for file mode.