Repository navigation
fix: bump prime-sandboxes to 0.5.3 and verifiers - #4368
Merged
Merged
Conversation
prime-sandboxes 0.5.3 stops capping live processes at 24h. A long agent segment on Prime no longer loses its ACP process to the SDK's stream deadline, which stopped the trace as harness_error instead of a timeout. The new verifiers requires prime-runs>=0.1.5, so take prime-runs 0.1.5 and prime-traces 0.0.11 from PyPI and drop their adf8d25c git pins. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
kcoopermiller
approved these changes
Nov 2, 2026
mikasenghaas
pushed a commit
that referenced
this pull request
Nov 2, 2026
## Summary - Raise the `prime` floor from `>=0.18.0` to `>=0.19.2` and relock (`uv.lock`: `prime` 0.18.0 → 0.19.2, `prime-sandboxes` 0.5.3 → 0.5.4). No other lock changes. ### CLI/SDK changes pulled in - **feat(sandboxes): volume size at create + grow** (PrimeIntellect-ai/prime#1132, released in PrimeIntellect-ai/prime#1144). `prime sandbox volumes create --size-gb N`, `prime sandbox volumes grow NAME --size-gb N`, and SDK `volumes.create(size_gb=)` / `volumes.grow()`. Additive only. - `prime` 0.19.0–0.19.1 (localhost forwards, pause/resume, 24h live-process fix) are already covered by `prime-sandboxes` 0.5.3 from #4368, which merged before this branch was cut — no overlap with it. Companion: PrimeIntellect-ai/verifiers#3045 (prime-sandboxes 0.5.4). ## Verification - `uv lock --check` passes (uv 0.12.3 and 0.13.6); `uv sync --all-extras --all-packages --locked` succeeds (resolves `prime` 0.19.2 + `prime-sandboxes` 0.5.4). - Standalone venv with 0.5.4: `Volume.size_gb` present; `Volumes.grow(name, size_gb)` exists. - `pytest tests/unit --ignore=tests/unit/train -m "not gpu"` (macOS): 737 passed, 1 skipped, 1 xfailed, 7 failed. The 7 failures (inference/orchestrator/transports) reproduce identically on `origin/main` in the same env — pre-existing, unrelated. 5 modules skipped at collection (missing linux-only `vllm`/`fla`/`prime_kernels`; CI runs on ubuntu).
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
prime-sandboxes0.5.1 → 0.5.3 so live sandbox processes are no longer capped at 24h (via fix: bump prime-sandboxes to 0.5.3 verifiers#3043).deps/verifiers60290d4→bccc2c5(verifiers main, with fix: bump prime-sandboxes to 0.5.3 verifiers#3043 merged).prime-runs[train]>=0.1.5from PyPI. This drops theprime-runsgit source and theprime-tracesgit override, both pinned toadf8d25c. The new verifiers requiresprime-runs>=0.1.5(chore: require prime-runs 0.1.5 from PyPI verifiers#3001), and the old git pin no longer resolves. 0.1.5 containsadf8d25c, so it includes perf: encode json with orjson in prime-runs and prime-traces prime#1109.uv.lock:prime-runs0.1.4 (git) → 0.1.5,prime-traces0.0.10 (git) → 0.0.11.Why
On Prime, the SDK sent a fixed 24h deadline on every live process stream, and the sandbox ended the process when it expired. A long agent segment then lost its ACP process. The trace stopped as
harness_error(EOFError: ACP process closed its stdout) instead ofagent_timeout, soagent_timeoutundercounted real 24h cap hits.prime-sandboxes changes (0.5.1 → 0.5.3)
0.5.3:
0.5.2:
verifiers changes (
60290d4→bccc2c5)Fixes and features:
Dependency pins:
Harness versions:
Dependabot lockfile bumps in verifiers (they do not change prime-rl's lock):
Verification
uv sync --frozen --all-packages --all-extrasinstallsprime-sandboxes0.5.3.prime_sandboxes.sandbox._LIVE_PROCESS_NO_TIMEOUT_MS == 9999999999is the default deadline for everyopen_processstream.uv.lockchanges onlyprime-sandboxes,prime-runsandprime-traces(plus a uv-written reorder of theexclude-newer-packagetable).pytest tests/unit --ignore=tests/unit/train: 837 passed, 1 xfailed.tests/unit/train(RTX PRO 6000 Blackwell, SM120):qwen3_8_flash_nextandtest_fp8_grouped_gemmshow the same failures, andtest_fused_lm_head.py::test_inject_prime_lm_head_vanillakills the process, on this branch and onmain(823ed27). These GPU kernel tests do not touch this diff.tests/unit/test_runtimes.py -k prime, includingtest_execon a live process) pass on 0.5.3.🤖 Generated with Claude Code