A collection of Claude Code agent skills for specialized tasks.
| Skill | Description |
|---|---|
| ck-kernel-resource-usage | Analyze GPU kernel resource usage (VGPRs, AGPRs, SGPRs, occupancy, scratch memory, LDS, register spills) by building a CK target with -Rpass-analysis=kernel-resource-usage and parsing the build log. |
| ck-ci-build | Trigger Jenkins CI builds for Composable Kernel PRs with optional parameter overrides. Uses curl for status checks and playwright-cli for browser-based build triggering. Supports saved parameter presets, in-progress build detection, and dry-run mode. |
| ck-ci-report | Analyze CI build failures for Composable Kernel PRs on the Jenkins dashboard. Uses playwright-cli for browser automation and the Blue Ocean REST API for deep-dive stage log extraction. Handles success, failure, aborted, infra, and no-build cases. |
| ck-fmha-codegen-guide | Generate comprehensive FMHA kernel selection guides by dynamically analyzing CK codegen source code. Produces filename structure docs, field references, minimal kernel sets, and filter commands for generate.py. |
| ck-list-fmha-prs | List open pull requests from ROCm/rocm-libraries focused on fused multi-head attention (FMHA) kernels. |
| post-review | Validate and post inline PR comments from a prior /review output. Spawns a subagent to verify each claim against actual code, drops wrong items, formats confirmed findings as inline comment drafts, and optionally posts them to GitHub. Works with both PR and local branch reviews. |
| review-changes | Perform evidence-based static review across Git revisions, working trees, folders, patches, or artifact versions before human review or acceptance. |
| gpu-usage | Report GPU usage on shared servers - shows which processes occupy each GPU, VRAM percentage, GPU utilization, process owner, elapsed time, and Docker container name. |
| hip-kernel-team | Launch and manage a multi-agent team for HIP/GPU kernel development, debugging, and optimization. Configurable roles (Lead, Implementer, Profiler, Researcher) with optional recall-plugin integration and long-running context rotation. |
| monorepo-bridge | Bidirectional commit transfer between monorepos and standalone submodule repos. CLI tool with setup, split, export, sync, reset, and verify subcommands. Uses fetch + subtree merge for export, rebase for sync. Auto-detects config from existing remotes. |
| p4-config | Manage Perforce server configurations - switch between P4 servers, download files from depot paths, and create new server configs. Stores settings in .p4config.<name> files. |
| handoff | Compact the current conversation into a single self-contained handoff document so a fresh agent can resume the work. Captures not just what's next but the operating rules, decisions, workflow, and dead ends, then emits a copy-paste kickoff prompt for the next session. |
| context-watchdog | Monitor Claude Code agent context usage across tmux panes. Scans all Claude panes in the current tmux window and reports their context percentage. Supports one-shot scan and recurring cron monitor modes. |
| write-html-doc | Produce a single self-contained HTML technical document in a fixed house style (light theme, all IBM Plex Mono, sticky sidebar TOC with scroll-spy, tables/callouts/chips, no charts). Converts a Markdown/text/source file to HTML or writes it straight from the conversation; editorialize or faithful conversion; locked template with per-request overrides. |
| teams-unread | List unread Microsoft Teams 1:1/group chats, or pull recent message history from a specific chat, by reading the Teams desktop app's UI Automation accessibility tree - no Teams API or bot registration needed. Windows only. |
| message-discipline | Gate every outbound inter-agent message on whether it carries substantive value and whether sending is required right now, suppressing bare acknowledgments and status placeholders regardless of role or recipient count. |
Install these through the Claude Code skill system or link them into the Claude Code skills directory. review-changes also follows the portable Agent Skills format and can be installed through another harness's skill-discovery mechanism.