steven112163
UserMonorepo for Claude Code harness engineering with codex
Categories
Indexed Skills (8)
ck-profile
Profile a Composable Kernel (CK) build target two ways: static compile-time resource analysis (registers, occupancy ceiling, spills, scratch, LDS) and dynamic runtime profiling with rocprofv3 (timing, HBM traffic, cache, occupancy) averaged over N runs and an argument sweep, with a roofline-lite compute/memory/latency verdict. Use when the user wants to profile, benchmark, or find the bottleneck of a CK example/test binary; says "profile <target>", "is <target> compute- or memory-bound", "check register spills / occupancy", "benchmark fp16 vs bf16", "rocprof <target>", "static/dynamic profile", or "sweep -B / -prec and compare". The user selects one or more of six modes: static, dynamic, trace, cfg, depgraph, compute. Not for non-CK or CPU-only targets, or for correctness testing.
council
Fan out a question to Claude and GPT-5.6-sol in parallel, run up to 3 rounds of adversarial debate where each model challenges the other's weakest argument and updates its position, assess convergence after each round with a dedicated subagent, then synthesize and deliver a final answer weighted by argument quality — not consensus.
llm
Query an external LLM to get a second opinion, cross-check a claim, debate a design, or consult a different model. Use when the user asks to consult another model, or on your own initiative when a second perspective would strengthen your answer. Default model is gpt-5.6-sol (1M context).
multi-review
Use when reviewing a diff or pull request before merge and a single review pass is not enough. Triggers include "multi review", "multi-angle review", "review my changes", "review my diff", "review PR <n>", "deep review", or wanting a thorough independent review of local uncommitted/branch changes or a GitHub PR. Reports findings in the conversation; does not post to GitHub.
research
Use when answering research questions, exploring ideas, or investigating technical topics. Triggers include open-ended questions ("how should we approach X"), factual queries ("what is X"), comparative analysis ("compare X vs Y"), or any request for thorough investigation. Usable by both humans and agents. Caller selects mode or the skill auto-detects from intent.
survey
Use when surveying academic literature — producing a literature review, related-work synthesis, state-of-the-art overview, or comparison of papers on a topic. Triggers include "survey the literature on X", "review papers about Y", "compare these papers", "what's the state of the art in Z", "related work for <topic>". Handles both topic-driven discovery and a supplied set of papers.
create-pr
Use when creating or opening a pull request, or pushing branch work for review, in any git repository. Triggers include "create PR", "open PR", "submit PR", "make a PR", "raise a PR", or wanting to turn a branch into a reviewable pull request. Applies Composable Kernel (CK) team conventions automatically when the repo is detected as CK.
dev-team
Use for complex, open-ended, or performance-sensitive development tasks where several implementation approaches should be compared and the best chosen on evidence. Triggers include kernel development or optimization, performance tuning with a target to hit, work needing both correctness verification and profiling, comparing alternative designs, or any task too large for a single agent to carry alone. Not for small, well-scoped single-file changes.
Bio shown is the top-scored skill's repo description as a fallback — real GitHub bios land in a future update.