pave-init
claudecodexCreate a PAVE-based Claude Code workflow plugin (a lead skill plus registered role agents) from a stated goal, with adversarial review gates.
3 skills · 10 agents · hooks
pave-init-generated workflow: bring the vLLM-Neuron platform plugin fork to parity with upstream GPU vLLM through user-gated, evidence-backed campaigns. Ships the fifteen Neuron skills its campaign stages load inside the plugin (see its README).
Plugin vllm-neuron-parity from marketplace jinhuang12-plugins in jinhuang12/pave. Commands read from the repo's manifest — 8 days ago.
This plugin runs code on your machine — its hooks fire automatically on events like tool use. The exact commands are listed below; read them before installing.
experimental-neuron-autoport-compiler-debugging-vllm-neuronDebug BRINGUP / first-compile failures when porting a NEW model architecture to the vllm-neuron Trainium2 backend (native torch.compile route by default since 0.29, XLA route deprecated): a graph that will NOT trace, lower, verify, fit, schedule, or produce correct output — NOT tuning a model that a
experimental-neuron-framework-benchmark-vllmCross-platform comparative serving benchmark that PROVISIONS hardware (rents a GPU box / Kaizen-provisions Trainium) and runs `vllm bench serve` with one identical workload across stacks, so NVIDIA GPU (upstream vLLM) and Trainium (vLLM-Neuron) throughput/latency are directly comparable. Use ONLY fo
experimental-neuron-framework-optimizationEnd-to-end workflow for optimizing a model on AWS Neuron (Trainium/vLLM-Neuron) to meet a joint latency + throughput SLA. Delegates planning, entitlement, profiling, analysis, implementation and compiler debugging through the campaign controller, behind two decision gates: Gate A (classify the ask —
experimental-neuron-infra-runtime-troubleshootingDiagnose Neuron device errors, compilation failures, OOM, runtime crashes, and performance issues on Trainium/Inferentia.
experimental-neuron-perf-compiler-artifact-debuggingDebug performance issues in framework-compiled (vllm-neuron / torch-xla / jax) Neuron workloads via compiler artifacts: HLO (proto/text), penguin IR, per-pass git history, and compile logs. Use for: redundant DMA / must-alias bugs, compiler-version diffs, internal compiler crashes (NCC_IMPR901, NCC_
neuron-framework-autoport-vllm-neuronPort a HuggingFace model architecture to the vLLM-Neuron Trainium2 backend. Handles architecture research, code generation, and validation.
neuron-framework-equivalenceVerifies functional equivalence between two implementations of the same model using a hierarchical 8-stage algorithm. Orchestrates model tree comparison, component-level R-ratio testing, E2E logit comparison, and distributional checks. Use when porting models between frameworks, hardware targets, or
neuron-framework-profile-analysis-vllm-neuronIngests vllm-neuron model execution artifacts into parquet using neuron-explorer and guides profile analysis: identifies prefill/decode NEFFs, segments by collective boundaries, computes per-engine bounds and DMA efficiency, and produces structured workload reports. Use when encountering "analyze pr
neuron-framework-profiling-vllm-neuronGuides profiling vllm-neuron models on Trainium: capturing device (NTFF) and system traces from a live server via vLLM's /start_profile and /stop_profile endpoints (the primary method), or offline via `vllm bench throughput` + `neuron-explorer capture` (fallback). Covers the neuron_profiler config,
neuron-nki-debuggingThis skill guides debugging NKI compilation errors on Neuron hardware. Use when encountering "compiler error on device", "debug NKI kernel", "test kernel on trn2/trn3", "neuronx-cc compilation failed", "validate kernel on hardware", "run kernel on trainium", or asking "how to debug NKI compilation e
neuron-nki-docsResearch NKI documentation for API lookups, tutorials, error codes, architecture, and optimization guides. Use when user asks "what does <API> do", "how to do <task>", "what is error <code>", "NKI <API> signature", "find NKI tutorial for <topic>", "look up <symbol>", or needs any NKI documentation r
neuron-nki-profile-queryingQuery and analyze NKI kernel profile data from neuron-explorer parquet files. Supports SQL queries via the neuron-explorer API and Python on parquet for advanced analysis. Works locally on trainium with NEFF/NTFF files on disk. Querying: start neuron-explorer, ingest profiles, run SQL against tables
neuron-nki-profilingThis skill guides using the cli to generate NKI kernel profiles (NEFF + NTFF pairs) to analyze performance on Neuron hardware. Use when encountering "profile kernel", "capture execution trace", "generate NEFF", "get summary-json", or asking "how to profile NKI kernel".
neuron-nki-writingGuide for writing and modifying NKI kernels. Covers new kernel creation from PyTorch/NumPy/natural language, editing existing kernels, adding shape/dtype support, refactoring tiling strategies, and implementing new features in NKI code. Use when user says "write NKI kernel", "convert PyTorch to NKI"
neuron-parallelism-plannerPlans tensor/pipeline/context-parallelism (TP/PP/CP) and context-bucket configurations for transformer LLM inference on AWS Neuron (Trainium trn1/trn2/trn3) using an analytic hardware-entitlement model validated against a meta-device FLOP-counter oracle. Use when deciding how to shard a model for se
vllm-neuron-parityBring the vLLM-Neuron platform plugin fork to parity with upstream GPU vLLM: scan the upstream delta, cost and rank each target's closing route, then run user-gated campaigns through correctness and performance gates against a GPU baseline into evidence-backed fork PRs. Manual-only: use only when th
adjudicatorIssues vllm-neuron-parity gate verdicts from stable, pre-registered evidence and names their consequence; never the agent that produced the evidence. Dispatched by the vllm-neuron-parity lead only — do not trigger from an implicit match.
adversarial-reviewerMaterial-findings-only adversarial review of vllm-neuron-parity's high-stakes artifacts — route verdicts, campaign designs, kernel increments, the whole-campaign code and PR description — independent of each artifact's producer. Dispatched by the vllm-neuron-parity lead only — do not trigger from an
implementerDesigns and builds vllm-neuron-parity campaign candidates in isolated worktrees, runs hardware attempts under lease, and writes the PR description. Dispatched by the vllm-neuron-parity lead only — do not trigger from an implicit match.
investigatorGap scan, route analysis, costing, and design-entry triage for a vllm-neuron-parity run — gathers and weighs evidence, never approves results. Dispatched by the vllm-neuron-parity lead only — do not trigger from an implicit match.
measurerExecutes vllm-neuron-parity's hardware calibration suite and pre-registered measurement procedures against the GPU baseline and writes stable evidence artifacts; never adjudicates. Dispatched by the vllm-neuron-parity lead only — do not trigger from an implicit match.
rederiverFirst-principles re-derivation for a vllm-neuron-parity campaign after a stop limit trips, a measurement path dead-ends, a repair loop reports no progress, or the user declines a close-out. Read-only on inputs; its output re-enters design. Dispatched by the vllm-neuron-parity lead only — do not trig
python3 "${CLAUDE_PLUGIN_ROOT}/hooks/pre_tool_use_router.py" protected-branch
python3 "${CLAUDE_PLUGIN_ROOT}/hooks/pre_tool_use_router.py" compile-cache
python3 "${CLAUDE_PLUGIN_ROOT}/hooks/pre_tool_use_router.py" venv-opt
python3 "${CLAUDE_PLUGIN_ROOT}/hooks/pre_tool_use_router.py" write-limits
python3 "${CLAUDE_PLUGIN_ROOT}/hooks/pre_tool_use_router.py" no-retry-copy
python3 "${CLAUDE_PLUGIN_ROOT}/hooks/pre_tool_use_router.py" audit-sidecar
python3 "${CLAUDE_PLUGIN_ROOT}/hooks/dispatch_advisory.py"
"${CLAUDE_PLUGIN_ROOT}/skills/vllm-neuron-parity/hooks/graph_edit_guard.sh"
python3 "${CLAUDE_PLUGIN_ROOT}/hooks/pre_tool_use_router.py" write-limits
python3 "${CLAUDE_PLUGIN_ROOT}/hooks/pre_tool_use_router.py" no-retry-copy
python3 "${CLAUDE_PLUGIN_ROOT}/hooks/pre_tool_use_router.py" audit-sidecar
"${CLAUDE_PLUGIN_ROOT}/skills/vllm-neuron-parity/hooks/state-staleness-reminder.sh"
python3 "${CLAUDE_PLUGIN_ROOT}/hooks/write_log.py"
"${CLAUDE_PLUGIN_ROOT}/skills/vllm-neuron-parity/hooks/write-for-reader.sh"
python3 "${CLAUDE_PLUGIN_ROOT}/hooks/pre_tool_use_router.py" audit-hook-record
"${CLAUDE_PLUGIN_ROOT}/skills/vllm-neuron-parity/hooks/goal-restate.sh"
"${CLAUDE_PLUGIN_ROOT}/skills/vllm-neuron-parity/hooks/goal-restate.sh"
"${CLAUDE_PLUGIN_ROOT}/skills/vllm-neuron-parity/hooks/stop-guard.sh"
Create a PAVE-based Claude Code workflow plugin (a lead skill plus registered role agents) from a stated goal, with adversarial review gates.
3 skills · 10 agents · hooks
Hand-ported PAVE workflow: run a staged, multi-agent campaign that profiles and optimizes Pumice inference latency on your own NVIDIA GPU host, inside a local Pumice checkout, on Claude Code or Codex (see its README).
2 skills · 8 agents · hooks
This page updates itself from your repo. Add the badge to your README, or get it Featured to pin it above the organic results with a followed link to your site.