## TL;DR
Download artifacts in parts with a resumable plan: Vertex AI experiment artifacts are large and the agent download timed out as one operation, so list the artifacts, download them one at a time to durable storage, and verify each before continuing. One giant download is one giant failure surface. Per-artifact downloads turn the same job into a checklist where a timeout costs minutes, not the whole transfer.

```text
agent timed out when agent tried to download Vertex AI experiment artifacts error timeout
```

## Use this when
- The check or run fails with this symptom: an agent times out downloading Vertex AI experiment artifacts.
- You are working on an agent workflow collecting Vertex AI experiment artifacts for comparison or archiving and the failure repeats on the same data or configuration.
- You need the failing run, call, or stage isolated before changing code or thresholds.

## Not for this skill when
- A single small artifact that still times out, which points at network or permission trouble rather than transfer size.
- The error text names a different tool, stage, or root cause. Match the verbatim error first and use that skill.

## Steps
1. List the artifacts for the experiment with their sizes. Verify: the download set is enumerated and sized
2. Download the largest artifact alone with progress visible. Verify: it completes, or its rate shows the realistic time for the set
3. Download the rest one by one, storing each in durable storage with its checksum. Verify: every stored artifact verifies
4. Record a manifest of what was downloaded. Verify: an interrupted session resumes from the manifest
5. Compare only from the downloaded set once the manifest is complete. Verify: the comparison uses verified artifacts

## Variant phrasings
### agent timed out download vertex ai experiment artifacts
The shorter phrasing focused on the named failure.
### vertex artifact download timeout agent error
The symptom phrasing a searcher tries next.
### vertex artifacts per file download fix
The variant naming the failing stage or tool directly.

Compatibility: Vertex AI experiment artifacts and model registry downloads, agent transfer workflows

## Why it happens
Experiment artifacts include full model binaries and evaluation outputs that dwarf the metadata an agent usually moves. A bulk download launched inside an agent session inherits the session limit, and cloud storage throttling stretches it further, so the transfer dies as a unit.

## Edge cases / pitfalls
- Session storage is often wiped between agent sessions; durable storage is part of the fix, not an optimisation.
- Checksum verification per artifact catches the silent truncation a timeout can leave behind.

## Provenance

Resolved from the public thread: https://vectle.com/posts/pst_ZprN5jd29-as62M2INCZXQ