prime images transfer-bulk

[kcoopermiller]

Adds a bulk counterpart to prime images push --source-image: copy many existing registry images into prime in one command

prime images transfer-bulk --manifest transfers.jsonl
prime images transfer-bulk --harbor ./tasks
prime images transfer-bulk --hf R2E-Gym/R2E-Gym-Subset --column docker_image

• --manifest — JSONL, one transfer per line: {"source": "ghcr.io/org/app:v1", "image"?: "name:tag", "platform"?: "..."}.
• --harbor — transfers each task's prebuilt [environment] docker_image. This is the exact complement of push-bulk's Harbor mode (which skips those tasks), so the two commands together cover a full task directory.
• --hf — reads image references straight from a Hugging Face dataset column via the dataset-viewer API (no datasets dependency, no download). Takes --column and supports --hf-config/--hf-split, dataset URLs, and HF_TOKEN for gated datasets.

How it runs

1. Validate everything up front like push-bulk. --dry-run prints the resolved source → destination table without touching the API.
2. Submit one transfer per request through the sliding-window logic shared with push-bulk (run_bulk_jobs refactored out of run_bulk_push). The server's transfer limiter counts images/minute (10 personal / 60 team), so on a plain 429 the spec is requeued and submission pauses briefly while in-flight transfers keep getting polled. A quota 429 stops submission (rest marked SKIPPED); 20 consecutive deferrals with zero progress makes the run give up instead of pacing forever.
3. Poll to terminal status and write anything that didn't complete to a re-runnable failures manifest: prime images transfer-bulk --manifest transfer-bulk-failures.jsonl.

Verified end-to-end against R2E-Gym/R2E-Gym-Subset (4,578 images resolved via dry-run).

---

Note

Medium Risk
Large CLI surface area touching image build/transfer APIs, quotas, and rate limits; platform-image team_id omission is a deliberate behavior change but scoped to admin platform flows.

Overview
Adds prime images transfer-bulk, the bulk counterpart to push --source-image, for copying many external registry images into Prime with the same operational shape as push-bulk (up-front validation, sliding-window concurrency, polling, re-runnable failure manifests).

transfer-bulk accepts exactly one of JSONL (source / optional image destination override), Harbor (each task’s prebuilt docker_image—complementing push-bulk’s build-from-environment/), or Hugging Face (--column image refs via the datasets-server API). It derives default Prime names/tags from source refs, treats transfer 429s as paced requeues (with a cap after repeated deferrals), and supports --platform-image like single-image push.

push-bulk gains --hf mode (Dockerfile text + name columns → temp build contexts) via new images_hf, and shares a refactored run_bulk_jobs engine with transfers: injectable submit, quota stop, rate-limit gating, server errorMessage on failed polls, and Ctrl+C resume via BulkRunInterrupted.

prime images push no longer errors on platform images in team context—it ignores team for platform transfer/build payloads and prints a dim notice instead.

^{Reviewed by Cursor Bugbot for commit 903fefa. Bugbot is set up for automated code reviews on this repo. Configure here.}

[cursor]

Cursor Bugbot has reviewed your changes and found 1 potential issue.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit 38aa296. Configure here.}

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 38aa296de6

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[kcoopermiller]

@codex review

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: cb1fc1cdea

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[kcoopermiller]

@codex review

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 16c4977749

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[kcoopermiller]

@codex review

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 16c4977749

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".