workspace

Workspace

Provides unified read/write access to all files in the current session. The userId and sessionId are injected automatically — you only specify the logical path.

Path Conventions

Prefix	What it accesses
uploads/<file>	Files uploaded by the user
code-agent/<file>	Files created by the code agent (auto-synced)
code-interpreter/<file>	Files created in the Code Interpreter workspace
documents/powerpoint/<file>	PowerPoint presentations
documents/word/<file>	Word documents
documents/excel/<file>	Excel spreadsheets
documents/image/<file>	Images from other tools

Note: Files written under /mnt/workspace persist for the chat session, and user uploads are available under /mnt/workspace/inputs. Code Agent downloads session uploads into its own working directory when delegated.

Usage

See everything in the session:

workspace_list()
workspace_list("uploads/")
workspace_list("code-agent/")

Read a file the code agent created:

workspace_read("uploads/data.jsonl")           # uploaded structured data
workspace_read("code-agent/calculator.png")   # binary → base64
workspace_read("code-agent/report.md")        # text → string

Pass a file from one skill to another:

result = workspace_read("documents/excel/data.xlsx")   # encoding: base64
workspace_write("code-agent/data.xlsx", result["content"], encoding="base64")

Notes

• Text files return encoding: "text" with plain string content. Large text reads are bounded; when truncated is true, use Code Interpreter for the complete file instead of repeatedly reading it into model context. Code Interpreter uses /mnt/workspace/inputs/<file>.
• JSON, JSONL, and NDJSON uploads are text files; large chat attachments may be truncated in model context, but the complete file remains under uploads/
• Binary files (images, Office docs, PDF, etc.) return encoding: "base64"
• workspace_write accepts both encodings — use "base64" for binary
• Files written here are immediately available to all other skills in the session