{"id":16570,"plugin_id":"plugins_6a57ef89f2d481918513e6133ec3fc18","kind":"skill","collection_source":null,"comparison_source":null,"observed_at":"2026-09-30T23:13:31.837Z","digest":"05a6c2991a57ed3d8b868d4d1c1e932bcd88e8c8b4c42cd87725064f7cb1f375","against":null,"payload":{"name":"tracelens-analysis-orchestrator","description":"Orchestrates modular PyTorch profiler trace analysis with TraceLens: generates perf reports, prepares category data, runs system-level and compute-kernel subagents in parallel, validates outputs, and writes a prioritized stakeholder report (analysis.md). Use when the user asks to follow the analysis orchestrator, run the agentic analysis workflow, analyze a trace, compare two traces, or mentions standalone or comparative TraceLens analysis.","included_files":[{"relative_path":".federated.json","size_in_bytes":268},{"relative_path":"agents/convolution-analyzer.md","size_in_bytes":7913},{"relative_path":"agents/cpu-idle-analyzer.md","size_in_bytes":6390},{"relative_path":"agents/elementwise-analyzer.md","size_in_bytes":7238},{"relative_path":"agents/gemm-analyzer.md","size_in_bytes":6409},{"relative_path":"agents/generic-op-analyzer.md","size_in_bytes":9727},{"relative_path":"agents/kernel-fusion-analyzer.md","size_in_bytes":15145},{"relative_path":"agents/model-identification-agent.md","size_in_bytes":4364},{"relative_path":"agents/moe-analyzer.md","size_in_bytes":8649},{"relative_path":"agents/multi-kernel-analyzer.md","size_in_bytes":12126},{"relative_path":"agents/norm-analyzer.md","size_in_bytes":6873},{"relative_path":"agents/reduce-analyzer.md","size_in_bytes":6453},{"relative_path":"agents/sdpa-analyzer.md","size_in_bytes":10880},{"relative_path":"agents/triton-analyzer.md","size_in_bytes":7013},{"relative_path":"evals/evals.json","size_in_bytes":2438},{"relative_path":"evals/hooks.py","size_in_bytes":6503},{"relative_path":"evals/machine.yml","size_in_bytes":12},{"relative_path":"reference.md","size_in_bytes":33244},{"relative_path":"skill-card.md","size_in_bytes":396},{"relative_path":"templates/analysis_template.md","size_in_bytes":25675},{"relative_path":"templates/sub_agent_spec.md","size_in_bytes":20803}],"skill_md_contents":"---\nname: tracelens-analysis-orchestrator\ndescription: >-\n  Orchestrates modular PyTorch profiler trace analysis with TraceLens: generates perf\n  reports, prepares category data, runs system-level and compute-kernel subagents in\n  parallel, validates outputs, and writes a prioritized stakeholder report (analysis.md).\n  Use when the user asks to follow the analysis orchestrator, run the agentic analysis\n  workflow, analyze a trace, compare two traces, or mentions standalone or comparative\n  TraceLens analysis.\nlicense: MIT\n---\n\n<!--\nCopyright (c) 2026 Advanced Micro Devices, Inc. All rights reserved.\n\nSee LICENSE for license information.\n-->\n\n# Analysis orchestrator\n\nCoordinate **system-level** analysis (CPU/idle, kernel fusion, multi-kernel / comm / memcpy) and **compute-kernel** analysis (GEMM, SDPA, elementwise, etc.): one trace load, shared prep, parallel subagents, then aggregation into `analysis.md`.\n\n## Full procedure\n\nFollow **[reference.md](reference.md)** for every step (user prompts, `<prefix>` / `{CMD}` usage, CLI commands, subagent launch text, validation, report `tee` order, plot embedding, and trace diagnostics).\n\n## Workflow index\n\n```\n0. Query User Inputs (Platform, Trace Path(s), Analysis Mode, Environment Setup)\n1. Generate Performance Report (branches on analysis mode: training vs inference then, comparison scope)\n2-5. Prepare Category Data (GPU Util, Top Ops, Tree Data, Multi-Kernel Data, Category Filtering)\n6. System-Level Analysis (PARALLEL) → system_findings/\n7. Compute Kernel Subagents (PARALLEL) → category_findings/\n   7.5. Aggregate → priority_data.json::findings[]\n8. Validate Subagent Outputs\n9. load_findings + Model Identification (subagent) → metadata/model_info.json\n10. Render performance PNG if agent_extension.py is absent\n11. Generate analysis.md (orchestrator writes via <prefix> tee), optional extension, embed PNG\n```\n\n## Rules\n\n- **Subagents:** Use the Task tool **only** where reference.md says “subagent” (Steps **6**, **7**, **9**). The orchestrator runs everything else, including Step 7.5, using the command prefix from `<output_dir>/cache/cmd_prefix.txt` (`{CMD}` substitution).\n- **Language:** Prefer vendor-agnostic terms (GPU kernels, collective communication, vendor GEMM library, DNN primitives, GPU graph). When quoting trace data, real kernel names are fine.\n- **Subagent prompts:** Point each subagent at the checked-in agent file under `TraceLens/Agent/Analysis/skills/analysis-orchestrator/agents/<name>.md` (see reference.md for exact paths and prompt shells).\n\n## Primary outputs\n\n- **Deliverable:** `<output_dir>/analysis.md`\n- **Internals:** `system_findings/`, `category_findings/`, `category_data/`, `metadata/`, `perf_report*.xlsx`, CSV folders — see package README for layout.\n\n## Agent layout\n\nProject subagents ship with this skill: `TraceLens/Agent/Analysis/skills/analysis-orchestrator/agents/*.md`.\n"},"changes":[],"summary":"First saved snapshot. No earlier version is available for comparison.","summary_kind":"deterministic","summary_metadata":{}}