[zoroyihan7]

• Description: PR-0 of the toolbox (multibranch W1) workstream's merge into main: the toolbox's small edits to modules that already exist on main, taken from multibranch-dev and kept only where the hunk is the toolbox's (git log on multibranch-dev, authored by zoroyihan7 / toolbox commits). Each hunk stands on main alone: no import of a module that is not on main. The later toolbox PRs (measure, runner, toolbox) import these names.
  • hyperloom_common/gpu_identity (moved there by refactor(common): give Hyperloom and KernelForge one shared package #2308): AMD_GPU_DEVICE_IDS, normalise_device_id, gpu_type_for_device_id (from 333d2dd; with the R9700/0x7551 correction siliangchen-amd made in e40ef60).
  • executors/__init__: exports of the names the measurement backend calls (grid runner, workload rendering and its option types, server lifecycle, subprocess kill and RoundWatch, launch evidence, prefix-cache and speculative-decoding refusals, TraceLens vLLM patch). The option-type exports came in with siliangchen-amd's sync 6f64999 for the same callers.
  • executors/_profile_setup: VLLM_PATCH_PROFILER_FIELDS, the two --profiler-config fields only a TraceLens-patched vLLM accepts, as one constant the renderer and the backend read (e81281d).
  • executors/_subprocess_kill: vLLM's Rust frontend ready line is a server-ready marker (22afe76).
  • executors/benchmark_result: harvest_leaked_artifacts(..., leak_root=) scans only the named root (f4b7589).
  • Left out on purpose: the outside_facts constant refactor of common/{env_safety,failure_signature,gpu_power_settings,provenance,visible_devices} (imports hyperloom.common.outside_facts, not on main; it goes with the rule-pack PR); HIP_RUNTIME_ERROR (already on main via fix(failure-signature): a HIP call that failed while running is not a missing kernel #2141); profile_{sglang,vllm,atom}.yaml, _grid_runner, executors/profile, bringup/ladder (no difference left from main); benchmark_backend/magpie_confined, _accuracy_gate, _magpie_patcher, _workload_envs, benchmark_config, magpie_agentx (other workstreams' hunks).
• Linked issue(s): none (multibranch merge plan, toolbox PR-0)
• Tests: added hyperloom_common/tests/test_gpu_identity.py, executors/tests/test_package_exports.py, test_harvest_named_leak_root_is_the_only_root_scanned, test_scan_server_log_increment_detects_vllm_rust_frontend_ready; the vLLM profile flag tests now also assert VLLM_PATCH_PROFILER_FIELDS is exactly what is added when patched and absent when not. Each new test fails against main's modules (ImportError / AttributeError / TypeError / AssertionError) and passes with this change. Ran on py3.10 and py3.11: pytest src/hyperloom/plugs/tests src/hyperloom/orchestrator/actions/executors/tests src/hyperloom/common/tests plus test_benchmark_result.py test_kill_spawned_server.py test_profile_and_kernel_handlers.py test_shared_profile_setup.py test_grid_runner_behavior_lock.py; ruff check, ruff format --check, lint-imports (11 kept, 0 broken).
• Size/complexity triggers crossed: none; no function added above the ceiling.
• If this simplifies or refactors: _apply_vllm_profile derives its two patched-only flags from VLLM_PATCH_PROFILER_FIELDS instead of two literal lines; same flags, pinned by the existing profile tests.
• Observable effect:
  • A vLLM server started with the Rust frontend (VLLM_USE_RUST_FRONTEND=1) prints no uvicorn line; it is now seen as ready from "OpenAI server is ready to accept requests". Before, the boot watchdog never saw it ready and ended the round at its boot share.
  • harvest_leaked_artifacts gains an optional leak_root; existing callers pass none and see no change. A caller that names its work directory no longer gets artifacts another test left in /workspace, $RESULT_DIR or INFERENCE_OPTIMIZER_LEAK_ROOTS copied into its output.
  • gpu_identity and the executors exports are new names with no caller on main yet; nothing an operator sees changes until the measurement PRs land.
• Breaking changes: no
• PR addresses single concern: no. It bundles the toolbox's small edits to existing main files so that the following toolbox PRs (new files only) have their imports on main; each item is listed above with its own effect.
• Root cause is upstream: no

🤖 Generated with Claude Code

[github-actions]

Code metrics gate: PASSED

Judged: the 10 files this change touches. Files it does not touch never fail it; anything found there is listed under Outside this change for information.

Dimension	Limit	New	Worse	Improved	Over the limit	vs base
Cyclomatic complexity	fails > 20	0	0	0	0	+0
Cognitive complexity	fails > 30	0	0	0	34	+0
Function length (lines)	fails > 60	0	0	0	249	+0
Nested block depth	fails > 4	0	0	0	37	+0
Parameter count	fails > 5	0	0	0	0	+0
Module length (lines)	warns > 800, fails > 1200	0	0	0	42	+0
Duplicated lines per file	fails > 0	0	0	0	76	+0
Dead code (vulture)	fails > 0	0	0	0	2	+0
Directory size (entries)	fails > 50	0	0	0	1	+0

Check	Findings	Effect
Comments	0
Broad catch without a reason	0
Broad catch that records nothing	0
English only	0
Production code imports test code	0
Repeated literals	0
A1 Private name of another package	0
A8 Flag argument	0
A9 Tuple return	0
A10 Raw vocabulary comparison	0
A11 Hidden global write	0
A12 Pure forwarder	0
A12 Single-implementation abstraction	0

Module length warning: over 800 lines, within the 1200-line failure limit (not failing): 2

Module	Lines	Limit
src/hyperloom/orchestrator/actions/executors/_subprocess_kill.py	1151	warns > 800
src/hyperloom/orchestrator/actions/executors/benchmark_result.py	924	warns > 800

Verdict computed by the gate scripts in /home/runner/work/_temp/code-metrics-gate, not by this tree's copy.

_{Thresholds and sources: [tool.hyperloom.code_metrics] in pyproject.toml. Tools: complexipy 8.0.1, jscpd 5.4.1, radon 6.0.1, ruff 0.16.2, vulture 2.16.}