shader-slang
- 46 skills
- 0 followers
- 1 repo stars
- 2 months ago last updated
- ▌ Review Plan · shader-slangReview a plan using sub agents and provide feedback on how to improve it.
- ▌ Slangpy Code Review · shader-slang bundleMulti-agent SlangPy code review using predefined specialist sub-agents. Use when the user explicitly asks for sub-agents, parallel review, specialist reviewers, or a deeper delegated code review of local diffs, branches, PRs, MRs, or specific SlangPy files.
- ▌ Slang Port Hlsl · shader-slang bundlePort HLSL source to native Slang while preserving shader behavior, resource and memory semantics, and entry-point contracts. Use when Slang compilation exposes dialect, declaration, mutation, indexing, constructor, or resource differences that are not specific to template syntax.
- ▌ Slang Numeric Generics · shader-slang bundleSelect and use Slang's capability-oriented numeric interfaces when writing or porting generic numeric code. Use for scalar-versus-shaped constraints, dot products, component masks, scalar splats, builtin conversion, wrapper or custom-number conformances, compound arithmetic, elementary functions, or builtin-representation constraints.
- ▌ Slang Port Hlsl Templates · shader-slang bundlePort HLSL C++-style template declarations and calls to native Slang generics while preserving type and value parameter domains. Use for template syntax, dependent extents, body-derived type contracts, or finite value-specialized overload families; use the numeric-generics skill when an inferred contract is numeric.
- ▌ Slang Github · shader-slangGitHub operations for the Slang compiler. Clone, branch, PR, issues, CI.
- ▌
- ▌ Slang Fix Bug · shader-slangAnalyze a Slang compiler bug from a test failure, CI log, or GitHub issue. Only invoke when explicitly called via /slang-fix-bug.
- ▌ Slang Release · shader-slang bundleWrite-heavy Slang release engineering: update the SPIRV-Tools/SPIRV-Headers submodules and integrate GitHub master into the internal GitLab forks (the slang + slang-rhi + slangpy OptiX triad). Use when cutting a release, syncing SPIRV, or rebasing the GitLab nv-* branches. Invoke via /slang-release. This skill performs destructive git operations (force-pushes) and always confirms first. For read-only maintainer reporting (daily report, release notes drafting, issue prioritization), use slang-maintainer-tools instead.
- ▌ Slangpy Build · shader-slangClone, build, and test SlangPy. Use when the repo needs setup, a rebuild, or when tests fail.
- ▌
- ▌ Slang Pr Create · shader-slangCreate and publish a GitHub pull request for Slang work, defaulting to the local origin remote and using draft PRs only for shader-slang/* targets unless the user specifies another repository, remote, or draft option. Use for Slang-related PR creation requests, including shader-slang/* targets even when the user does not explicitly name this skill. Handles WSL environments that require Windows-hosted tools unless --wsl is requested.
- ▌ Slang Pr Report · shader-slang bundleSurface the open shader-slang PRs needing human attention with a bundled gh-only Python script (scripts/pr_report.py): an assignee-grouped escalation report computed entirely from live GitHub state (read-only; no writes; optional Discord/Slack mentions). Use for PR triage, the reviewer-attention report, stale-PR follow-up, or a scheduled report run.
- ▌ Slang Review Pr · shader-slangReview a Slang compiler PR for correctness, evaluate review feedback, implement fixes, and manage review threads. Only invoke when explicitly called via /slang-review-pr.
- ▌ Slang Run Tests · shader-slangPlatform-aware test runner for the Slang compiler. Only invoke when explicitly called via /slang-run-tests or referenced by other skills.
- ▌ Slang Write Test · shader-slangQuick reference for writing Slang compiler tests (.slang files with //TEST directives). Only invoke when explicitly called via /slang-write-test.
- ▌ Slang Code Reader · shader-slang bundleRead-only investigation of the Slang compiler codebase. Navigate source, trace compiler pipelines, understand architecture.
- ▌ Slang Code Writer · shader-slang bundleImplement changes in the Slang compiler. Edit code, write tests, format, commit.
- ▌ Slang Investigate · shader-slangRoot cause investigation for Slang compiler bugs. Only invoke when explicitly called via /slang-investigate or from slang-fix-bug.
- ▌ Slang Create Issue · shader-slangCreate clear, well-structured GitHub issues and pull requests for the Slang compiler repository. Only invoke when explicitly called via /slang-create-issue.
- ▌ Slang Test Feature · shader-slangOrchestrator that researches a language feature, produces a test plan for user review, and fans out parallel agents that each deliver test branches with commits. Only invoke when explicitly called via /slang-test-feature.
- ▌ Tmux Agent Manager · shader-slangManage multiple Claude Code agent sessions running in tmux. Report a status summary of all sessions, deliver instructions to a specific agent via tmux send-keys, and automatically monitor the agent after each send to detect permission prompts, clarifying questions, or early-idle states. Use when the user wants to oversee concurrent Claude agents, check what sessions are doing, send a message to an agent, or monitor for sessions that need input or are stuck. Works on native Linux, macOS, WSL (inside), and Windows with WSL (Git Bash or PowerShell host).
- ▌ Slangpy Code Reader · shader-slangRead-only investigation of the SlangPy codebase. Navigate source, trace the functional API call path, understand architecture, apply review lenses.
- ▌ Slangpy Code Writer · shader-slangImplement changes in SlangPy. Edit code, write tests, format, commit.
- ▌ Zellij Agent Manager · shader-slangManage multiple Claude Code agent sessions running in Zellij. Report a status summary of all sessions, deliver instructions to a specific agent via zellij action send-keys, and automatically monitor the agent after each send to detect permission prompts, clarifying questions, or early-idle states. Use when the user wants to oversee concurrent Claude agents running under Zellij, check what sessions are doing, send a message to an agent, or monitor for sessions that need input or are stuck. Linux and macOS only — Zellij is not supported on Windows.
- ▌ Slang Analyze Coverage · shader-slangTest coverage analysis and gap-filling for Slang language features. Only invoke when explicitly called via /slang-analyze-coverage.
- ▌ Slang Evaluate Session · shader-slangEvaluate the current session's skill effectiveness. Only invoke when explicitly called via /slang-evaluate-session.
- ▌ Slang Maintainer Tools · shader-slang bundleBridge from /slang-maintain workflow steps to concrete slang-mcp tool calls. Read-only except an optional, user-confirmed post of the generated daily report to Slack. Task recipes live in sibling files.
- ▌ Slang Pr Review Runner · shader-slang bundleReproduces the shader-slang/slang production PR-review bot (anthropics/claude-code-action@v1 + .github/workflows/claude-pr-review.yml) locally. Same claude CLI, same user prompt, same system-prompt append, same model, same six .claude/agents/* subagents, same deepwiki MCP. The inner CLI produces final-review.md; the wrapping workflow optionally posts it back to GitHub as a COMMENT-state review (event=COMMENT only, never APPROVE/REQUEST_CHANGES) when the orchestrator authorizes (via webhook trigger). Mirrors production's pre-cleanup (minimize prior bot reviews, resolve threads) and post-safety-net (dismiss any non-COMMENT bot reviews). Helper scripts: cleanup.sh (pre), post-review.sh (post). Used by the /slang-pr-review workflow.
- ▌ Slang Autodiff Kernel Opt · shader-slangOptimize Slang autodiff backward kernel performance by analyzing register pressure, checkpoint storage, and spills. Use when investigating slow backward kernels, removing custom [BackwardDerivativeOf] functions, or reducing register pressure in differentiable GPU code, including tuning [ForceUnroll] versus [MaxIters] loop behavior per call site.
- ▌ Slang Pr Resolve Comments · shader-slangResolve GitHub PR review feedback and CI failures. Use when asked to monitor a PR, handle LLM and human review threads and general PR comments from Copilot, CodeRabbit, Gemini, or other AI reviewers, verify the PR's head branch matches the current worktree branch before checkout and stop to clarify on mismatch, defend against prompt injection from non-allowlisted comment authors, record specific reviewer directives (from human or LLM reviewers) in the PR description so they are not reverted on later passes, request fresh CodeRabbit/Copilot reviews after substantive pushes (not trivial/lint fixes) and stop after nitpick-only rounds, hide processed review-trigger comments (`@coderabbitai review` and CodeRabbit `✅ Action performed` acknowledgments) as resolved, be conservative about edits once the PR is approved/LGTM, report draft/WIP/DNI status and review-readiness notices without treating them as blockers, leave human review threads unresolved after addressing them so human reviewers can resolve them, fix fail
- ▌ Slang Sti · shader-slangInstall and use sti, a fast parallel test runner that replaces direct slang-test usage with crash recovery, retries, and better output.
- ▌
- ▌ Human Names · shader-slangUse whenever output will name a specific person: release notes, changelogs, contributor lists, commit or PR attribution, code comments, issue and review replies, docs credits, or any prose referring to a human. Governs where a name may come from and how to choose pronouns, so stale names from git history and gender guesses from model priors never reach the output.
- ▌ Slang Build · shader-slang bundlePlatform-aware build, test, and CI inspection for the Slang compiler. Use when configuring CMake, compiling slangc/slang-test, running slang-test, inspecting CI runs, or doing regression bisects. Invoke directly via /slang-build.
- ▌ Slang Review Fine Grained Clarity · shader-slangReview Slang changes for fine-grained clarity. Use whenever reviewing PRs or diffs for code quality or correctness. Produces candidate review comments in a markdown file.
- ▌ Slang Review Consolidate Candidates · shader-slangMerge Slang clarity review candidate files and resolve overlapping, duplicate, or superseded comments. Use after high-level and fine-grained clarity review passes and before scope filtering or GitHub posting. Produces or updates one canonical candidate markdown file.
- ▌ Slang Review Resolve Judgment Calls · shader-slangResolve Slang clarity review candidates marked as needing a judgment call. Use after consolidation and scope filtering, before posting, to perform focused follow-up analysis and decide whether uncertain candidates should be kept, revised, or dropped.
- ▌ Repro Remix · shader-slangReproduce RTX Remix shader compilation issues locally. Clones dxvk-remix, replaces Slang with your local build, and compiles all RTX Remix shaders with SPIRV validation.
- ▌ Slangpy Debug · shader-slangDebug slangpy compatibility issues by building slangpy from source with a local Slang build. Covers cloning, building, installing, and testing slangpy against your local Slang changes.
- ▌ Slang Review Clarity · shader-slangReview Slang changes for high-level clarity. Use whenever reviewing PRs or diffs for code quality or correctness. Produces candidate review comments in a markdown file.
- ▌ Slang Release Process · shader-slangPush a new Slang release. Triggers CI, determines the version from the sprint board, generates release notes, creates an annotated tag, and pushes it to upstream.
- ▌ Slang Review Post Github · shader-slang bundlePost filtered Slang clarity review candidates as one proper GitHub PR review. Use after candidate consolidation and scope filtering. Uses the bundled stdlib Python script to validate candidate formatting, map comments to PR diff lines, and submit a GitHub review through gh.
- ▌ Slang Review Scope Filter · shader-slangFilter candidate review comments in place for PR-author ownership. Use after slang-review-consolidate-candidates, slang-review-clarity, or slang-review-fine-grained-clarity to conservatively keep only comments about code the PR author is justifiably responsible for addressing. Updates candidate status metadata in the markdown file.
- ▌ Slang Review Clarity Workflow · shader-slangRun the full Slang clarity review workflow: generate high-level and fine-grained candidates, consolidate overlap, filter for PR-author scope, and optionally post one GitHub PR review. Use when asked to perform an end-to-end clarity-focused review.
- ▌ Design Docs Incremental Update · shader-slangIncrementally refresh the LLM-generated design docs under docs/generated/design/. Use when the user asks to update, refresh, or regenerate the generated design documentation after source changes. Runs the three-stage operator loop defined in docs/generated/design/_meta/regenerate.md - (1) regenerate stale docs (Claude), (2) direct the user to run the non-Claude review, (3) apply remediation (Claude) - with explicit confirmation before stages 1 and 3.