feat: add task download to save an agent's task artifacts to disk

[msampathkumar]

Goal: Let a user save an agent's file artifacts to local disk with one command, instead of only seeing them summarised in the terminal.
User story: As a developer whose agent produced files, I want task download to save the artifacts to disk, so that I can use the generated files directly instead of extracting base64 from JSON by hand.

Priority: P2 · Area: CLI code

Problem

The task command group covers get, list, subscribe, cancel, and push-config, but there is no way to save a task's file artifacts to disk. task get and send only render/report artifacts (and a file part currently shows as [binary N bytes]), so a caller who wants the actual bytes has no supported command.

task download does not exist today:

$ a2a task download --help
Work with tasks

Usage:
  a2a task [command]

Proposed behavior

Add a2a task download <taskId> that writes the task's artifacts to disk:

a2a task download <taskId> [--output-dir <dir>] [--artifact <id>]

• Writes each file artifact to --output-dir (default: current directory), using the artifact's name and media type; falls back to a safe generated name when none is provided.
• --artifact <id> (repeatable) downloads only the named artifact(s); default downloads all.
• Data/text artifacts are written too (e.g. .json/.txt), so nothing is silently skipped.
• Reports each path written; refuses to overwrite unless --force (or writes uniquely-named files).
• Works across transports and honours -o json (emit the list of written paths as structured output).

Who benefits

• Developers pulling generated files (reports, images, data) from an agent without hand-parsing JSON.
• CI jobs that archive an agent's outputs as build artifacts.
• AI coding agents that need the produced files on disk to continue a workflow.

Acceptance criteria

• a2a task download <taskId> writes all artifacts to disk and prints the paths.
• --output-dir and --artifact selection work.
• File artifacts use name + media type; collisions handled safely.
• -o json emits the written paths as structured output; exit code reflects success/failure.
• Tests cover: all-artifacts download, single --artifact, and a missing task.

Environment

• CLI version: a2a version v0.0.0-…1e29dfe94f95+dirty
• OS: Linux x86_64

Notes

task list, task subscribe, and task push-config already exist — this request is scoped to the one missing member of the group, task download. The rendering of file parts by name/media-type/size is tracked separately and complements this command.

[yarolegovich]

Artifacts are not files.

artifact's name and media type;

They don't have media type.

--artifact (repeatable) downloads only the named artifact(s); default downloads all.

Is a bit misleading, because we'll always download all, artifacts are not individually addressable in the protocol.

An artifact can be a full agent response containing thinking, message and an inline file. An Artifact wire representation would have base64 encoded bytes and I would expect a2a task download --artifact <id> to store it as a file, so the claimed problem is not addressed:

As a developer whose agent produced files, I want task download to save the artifacts to disk, so that I can use the generated files directly instead of extracting base64 from JSON by hand.

Is not addresses. You'll just have a file from which you'll need to extract base64 bytes.

In terms of command surface consistency I'd also rather have:

a2a task get <task_id> --artifact <id> 

To fetch the task and print a single artifact. You can then save it using shell redirects:

a2a task get <task_id> --artifact <id> -o json > artifact.json

But I agree it's a real problem and I've actually just opened another issue for it: #41