Agent skill

Replay Nsight

by Luna5ama in Luna5ama/Alpha-Piscium

Use glc2vk replayers and Nsight GPU Trace for offline Alpha-Piscium diagnostics after Vibris MCP captures.

GPL-3.0Auto-check passed

Install Replay Nsight

skills CLI
$ npx skills add Luna5ama/Alpha-Piscium --skill replay-nsight -a claude-code

Project install by default; add -g for ~/.claude/skills/.

GitHub CLI
$ gh skill install Luna5ama/Alpha-Piscium replay-nsight --agent claude-code

Project scope by default; add --scope user for a personal install. Needs GitHub CLI 2.90.0 or later (public preview).

Manual copy
$ git clone --depth 1 https://github.com/Luna5ama/Alpha-Piscium.git skills-src && mkdir -p .claude/skills && cp -r skills-src/.agents/skills/replay-nsight .claude/skills/replay-nsight && rm -rf skills-src

Use ~/.claude/skills/ instead of .claude/skills for a personal install. The folder must contain SKILL.md.

Claude Code skills documentation · loads skills from .claude/skills/

Facts

Skill name
replay-nsight
GitHub stars
156
Token cost
~2.1k tokens
SKILL.md length
1,018 words
Files
9 (incl. scripts)
Skills in repo
3
Repo updated
First seen
Licence
GPL-3.0

At a glance

Use glc2vk replayers and Nsight GPU Trace for offline Alpha-Piscium diagnostics after Vibris MCP captures.

  • Works in 2 steps: an exact capture directory when it… → otherwise, a root directory containing…
  • SKILL.md covers Files, Live control, Replayer Usage and Nsight GPU Trace, plus 1 more section
  • Runs PowerShell scripts from its folder

What it does

Replay Nsight is an agent skill from Luna5ama/Alpha-Piscium. Use glc2vk replayers and Nsight GPU Trace for offline Alpha-Piscium diagnostics after Vibris MCP captures.

Its SKILL.md is about 2.1k tokens, which your agent loads only when the skill is triggered. The skill folder holds 11 other files, including scripts (for example `config.example.json`).

It works with Model Context Protocol. The repository describes itself as: High-quality realistic Minecraft shaderpack. The licence is GPL-3.0.

Example prompts

  • “/replay-nsight”

Requirements

  • PowerShell

Workflow steps

2 steps, taken from the first numbered list in SKILL.md.

  1. an exact capture directory when it contains resource_metadata.json;
  2. otherwise, a root directory containing capture subdirectories, selecting the

What it can do on your machine

Read from SKILL.md and the folder at commit f2401e9. It shows what the files ask for, not the result of running them.

  • Tool permissions

    Pre-approves nothing: there is no allowed-tools line, so your agent's usual permission prompts apply.

    From allowed-tools in the SKILL.md frontmatter.

  • Runs code

    Ships 3 files in scripts/ (PowerShell), which the agent can run.

    From the folder's file list and the shell code blocks in SKILL.md.

  • Network

    No URLs in SKILL.md.

    From URLs in SKILL.md, links to its own repository left out.

  • Credentials

    Names no API keys, tokens, secrets or passwords.

    From names ending in _API_KEY, _TOKEN, _SECRET, _KEY or _PASSWORD in SKILL.md.

Context cost

Replay Nsight loads about 2.1k tokens when it runs. Until then it costs about 30 tokens; SKILL.md has 1,018 words of instructions outside code blocks.

Always · name and description, kept in context so the agent knows when to use it
~30
When it runs · the whole SKILL.md, loaded when a task matches
~2.1k

Estimates: characters ÷ 4, the usual rule of thumb; real counts depend on the model's tokenizer. Scripts and assets cost tokens only if the agent reads them.

Safety

Auto-check passed

The automated check found no risky patterns in SKILL.md.

Automated static check — not a guarantee. Review scripts before installing. It scans the text of SKILL.md for risky patterns (piping downloads into a shell, reading credential files, hidden Unicode, destructive commands); the scripts in this folder are not scanned.

SKILL.md

The full file from Luna5ama/Alpha-Piscium at commit f2401e9, republished under its GPL-3.0 licence (© Luna5ama). 1,018 words, ~2,119 tokens.

Download SKILL.mdSave it as .claude/skills/replay-nsight/SKILL.md (or your agent's skills folder). This skill also uses 8 other files; get the full folder from GitHub.
name
replay-nsight
description
Use glc2vk replayers and Nsight GPU Trace for offline Alpha-Piscium diagnostics after Vibris MCP captures.

Replay Nsight

Use Vibris MCP directly for shader reloads, live captures, and performance profiles. Use this skill only for offline glc2vk replay or Nsight GPU Trace analysis of a capture produced by MCP.

Files

  • bin/replay-gl.jar: OpenGL replayer.
  • bin/replay-vk.jar: Vulkan replayer.
  • config.json: local machine defaults. This file is ignored by git.
  • config.example.json: portable template.
  • scripts/run-replayer.ps1: runs the OpenGL or Vulkan replayer.
  • scripts/capture-gputrace.ps1: runs a replayer under Nsight GPU Trace.
  • <project-root>/.tmp/vibris/: generated replay argfiles and replay AOT caches.

Always prefer config.json for jdk and capture_path. If a command does not pass an explicit capture path, the scripts resolve capture_path as:

  1. an exact capture directory when it contains resource_metadata.json;
  2. otherwise, a root directory containing capture subdirectories, selecting the newest resource_metadata.json.

replay_frames controls the default frame count for both replayer scripts. Prefer running several frames rather than a single frame; this gives the driver and replay setup room to settle before the measured frame.

Both replay scripts write generated argfiles and replay AOT caches under the project root at .tmp/vibris/, not under the skill directory. The current files are replay-<backend>.args and replay-<backend>.aot.

Live control

Use vibris_profile for routine performance comparisons. It snapshots the worktree, applies shader config, reloads, resets temporal state, warms up, and measures the requested future frames in the running game. Use vibris_capture_pass for a known compute pass or vibris_capture_multi for a prepare, begin, deferred, or composite group. Poll vibris_get_capture_status until capture and saving complete, then pass its output path to the replayers.

Captured resources are recorded on first reference. If replacement shader code uses a resource absent from the capture, recapture instead of accepting an invalid replay.

Replayer Usage

Use scripts/run-replayer.ps1 with -Backend gl or -Backend vk. Omit -Capture to use config.json capture selection, or pass a capture directory explicitly. Use -ShaderRoot <path> for replacement shaderpack source and -ShaderPass <passName> to limit replacement to selected passes.

The underlying replay CLI accepts <captureDir> [frameLimit] [--shader-path|--shader-root <path>] [--shader-pass <passName>]....

OpenGL replay uses the captured original OpenGL shader source unless --shader-root is supplied. Vulkan replay uses runtime patching for shader replacement. Do not manually copy edited shaderpack sources into the capture directory; pass -ShaderRoot instead. When using a full shaderpack root to test a known hotspot, also pass -ShaderPass <passName> so only that pass is replaced and the other captured passes keep their original shader sources.

Nsight GPU Trace

Use the bundled Nsight Analyzer at ..\nsight-graphics-analyzer\scripts\nsight.py. Do not read BASE/GPUTRACE_REGIMES.xls directly; use the generated JSON files or the gputrace-stages, gputrace-actions, and gputrace-metric analyzer subcommands.

In most cases, capture GPU Trace with the OpenGL replayer. It uses the same OpenGL shader compiler path as the target game and is the best default for Alpha-Piscium performance work.

Use the Vulkan replayer only when you need finer Nsight information that is harder to get from OpenGL, such as shader source profiling or per-line stall reason detail inside one pass. In that case, run Vulkan trace with Nsight multi-pass metrics and shader profiling options. Treat the result as a diagnostic lens, not as exact OpenGL performance truth: Vulkan uses a different driver compiler path, so shader codegen and stall behavior may differ from the OpenGL shaderpack runtime.

Use scripts/capture-gputrace.ps1 with -Backend gl by default. Select a capture with -Capture, replacement source with -ShaderRoot and -ShaderPass, and trace timing or limits with the parameters below. For fine-grained Vulkan diagnostics, use -Backend vk with -MultiPassMetrics and -ShaderProfile.

GPU Trace parameter coverage in capture-gputrace.ps1:

  • Start trigger: -StartAfterFrames, -StartAfterSubmits, -StartAfterMs, -StartAfterHotkey, -StartWithNgfxSdk, -StartOnReplayBegin.
  • Stop/limit: -MaxDurationMs, -LimitToFrames, -LimitToSubmits, -StopWithNgfxSdk, -StopOnReplayEnd.
  • Buffers: -AllocatedEventBufferMemoryKb, -AllocatedHesBufferMemoryKb, -AllocatedTimestamps.
  • Metrics: -Architecture, -MetricSet, -MetricSetId, -PerArchConfigPath, -MultiPassMetrics.
  • Profiling: -NoTimeEveryAction, -ShaderProfile, -PerLineActiveThreads, -PcSamplesPerPmIntervalPerSm, -PmBandwidthLimit, -HesEnabled.
  • Collection and runtime: -SetGpuClocks, -CollectScreenshot, -DisableCollectShaderPipelines, -DisableCollectExternalShaderDebugInfo, -DisableTraceShaderBindings, -DisableNvtxRanges, -AllowTracingReplayReset, -KeepGoing, -TraceTimeout, -UseNgfxTimeout, -VerboseNgfx, -Timeout, -DryRun.
Show full SKILL.md (406 more words)Show less

All GPU Trace defaults live in config.json under gpu_trace_args. Command line parameters override JSON only when explicitly provided; otherwise capture-gputrace.ps1 reads the matching JSON key. This makes the trace setup editable by both users and agents without changing the script.

For GPU Trace, prefer start_after_frames instead of a millisecond delay. The default is start_after_frames = 3, which skips the first few replay frames and filters out initialization noise. Keep replay_frames larger than the start frame plus the capture limit.

When multi_pass_metrics is enabled and -Frames is not explicitly passed, capture-gputrace.ps1 uses gpu_trace_args.multi_pass_replay_frames, default 1000. This keeps the replay process alive long enough for Nsight multi-pass collection; the default process behavior is to exit after the GPU Trace capture finishes.

Some raw ngfx.exe --help-all options are not exposed by the current Nsight analyzer wrapper. Prefer the wrapper for normal work because it auto-exports the trace and writes the summary/stages/actions JSON files.

Successful Nsight GPU Trace output includes:

  • <name>.ngfx-gputrace
  • <name>.gputrace.summary.json
  • <name>.gputrace.stages.json
  • <name>.gputrace.actions.json
  • BASE/*.xls
Replayer trace measurement contract

For replay-gl.jar and replay-vk.jar traces, use summary.json only to confirm bundle_complete=True. Never use whole-capture duration, relative-to-capture values, analysis.frame_budget, fraction_of_gpu, replayer CPU submission, replay-perf, Copy work, or sleep/yield time as shader performance evidence. Those values include replayer behavior and unrelated scheduling noise that do not represent Iris runtime performance.

Report only:

  • the duration of an individual pass inside the outer Replay marker; or
  • that pass duration divided by the complete outer Replay marker duration.

Prefer gputrace-actions <trace> --in-marker "^Replay$" to select passes, and use the stage tree to read the exact outer Replay duration. Exclude the outer Copy marker and the unmarked tail sentinel dispatch from shader comparisons. The sentinel exists only to absorb end-of-replay timing noise.

  1. Ensure Minecraft/Iris is running in the target scene.
  2. Run vibris_profile for normal in-game performance comparisons.
  3. For offline diagnostics, capture only the target pass for a known hotspot; otherwise capture the relevant composite-like program type.
  4. Wait for vibris_get_capture_status to report that capture and saving are complete.
  5. Verify replay correctness with the OpenGL replayer.
  6. Collect GPU Trace with the OpenGL replayer unless Vulkan-only diagnostics are required.
  7. Confirm bundle_complete=True, then inspect only passes inside the outer Replay marker under the measurement contract above.

For shader experiments, edit shaderpack sources and pass -ShaderRoot; do not recapture unless resource or uniform usage changed. If new shader code uses a resource that was not referenced during capture, recapture that pass or program type.

© Luna5ama, GPL-3.0. Rendered from Markdown: HTML in the file is shown as text, images as links, and headings moved down two levels. Raw file

Files

SKILL.md and 8 other files (scripts) in .agents/skills/replay-nsight of Luna5ama/Alpha-Piscium.

  • SKILL.md
  • .gitignore
  • bin/replay-gl.jar
  • bin/replay-vk.jar
  • config.example.json
  • scripts/capture-gputrace.ps1
  • scripts/common.ps1
  • scripts/run-replayer.ps1
  • tests/test-aot-cache.ps1

Open the folder on GitHubat commit f2401e9

Compare with similar skills

Replay Nsight next to the 5 skills that share the most tags, products or categories with it. Stars are the repository's; “used in” counts other GitHub owners with a copy.

Replay Nsight compared with similar skills
SkillStarsUsed inTokensAuto-checkLicenceRepo updated
Replay Nsight this skillLuna5ama/Alpha-Piscium156—~2.1kAutomated safety check: PassGPL-3.0
MCP Server Builderanthropics/skills180k63 repos~2.3kAutomated safety check: PassApache-2.0
MCP Server BuildershareAI-lab/learn-claude-code78k4 repos~1.2kAutomated safety check: PassMIT
MCP Integration for Pluginsanthropics/claude-plugins-official38k11 repos~3.1kAutomated safety check: PassApache-2.0
Figma use_figma Plugin API Ruleswarpdotdev/warp65k4 repos~4.4kAutomated safety check: PassAGPL-3.0
Stitch to Remotion Walkthrough Videosgoogle-labs-code/stitch-skills8.5k6 repos~3.2kAutomated safety check: NotesApache-2.0

Similar skills

  • MCP Server Builder

    anthropics/skills

    Official

    Guides the design and implementation of Model Context Protocol servers in TypeScript or Python, from tool naming and error messages to evaluation.

    180k GitHub starsUsed in 63 repos~2.3k tokens
    Agent WorkflowsAuto-check passed
  • MCP Server Builder

    shareAI-lab/learn-claude-code

    Walks through building MCP servers in Python or TypeScript that expose tools, resources and prompts to Claude, with templates, registration and testing.

    78k GitHub starsUsed in 4 repos~1.2k tokens
    Agent WorkflowsAuto-check passed
  • MCP Integration for Plugins

    anthropics/claude-plugins-official

    Official

    Explains how to bundle Model Context Protocol servers in a Claude Code plugin, covering config files, stdio, SSE, HTTP and WebSocket server types, and authentication.

    38k GitHub starsUsed in 11 repos~3.1k tokens
    Agent WorkflowsAuto-check passed
  • Required groundwork before any use_figma call: the rules and reference files for running JavaScript in a Figma file through the Plugin API without common failures.

    65k GitHub starsUsed in 4 repos~4.4k tokens
    Frontend & DesignAuto-check passed
  • Stitch to Remotion Walkthrough Videos

    google-labs-code/stitch-skills

    Official

    Builds walkthrough videos from Stitch design projects using Remotion, with transitions, zoom effects and text overlays on each screen.

    8.5k GitHub starsUsed in 6 repos~3.2k tokens
    Media & CreativeAuto-check: notes
  • MCP Development

    coollabsio/coolify

    A skill your agent uses for Laravel MCP development. An agent skill from coollabsio/coolify.

    63k GitHub starsUsed in 1 repo~949 tokens
    Frontend & DesignAuto-check passed

More from Luna5ama/Alpha-Piscium

  • Nsight Graphics Analyzer

    Luna5ama/Alpha-Piscium

    Drive NVIDIA Nsight Graphics 2026.1+ from the command line for GPU performance analysis, frame capture, frame trace inspection, draw-call inspection, NVTX/D3DPERF stage timing, replay metadata…

    156 GitHub stars~4.7k tokensUpdated 6 days ago
    Auto-check passed
  • Write Alpha Piscium Changelog

    Luna5ama/Alpha-Piscium

    Draft, revise, translate, and finalize Alpha Piscium stable, beta, alpha, and hotfix changelogs from Git history.

    156 GitHub stars~1.9k tokensUpdated 6 days ago
    Auto-check passed

Questions about Replay Nsight

What does Replay Nsight do?

Use glc2vk replayers and Nsight GPU Trace for offline Alpha-Piscium diagnostics after Vibris MCP captures. Replay Nsight is an agent skill from Luna5ama/Alpha-Piscium. Use glc2vk replayers and Nsight GPU Trace for offline Alpha-Piscium diagnostics after Vibris MCP captures.

How do I install Replay Nsight in Claude Code?

Run `npx skills add Luna5ama/Alpha-Piscium --skill replay-nsight -a claude-code`. Or copy the skill folder (.agents/skills/replay-nsight in Luna5ama/Alpha-Piscium) into .claude/skills/replay-nsight in your project. Claude Code loads it when a task matches its description.

How do I install Replay Nsight in Codex?

Run `npx skills add Luna5ama/Alpha-Piscium --skill replay-nsight -a codex`. Or copy the skill folder (.agents/skills/replay-nsight in Luna5ama/Alpha-Piscium) into .agents/skills/replay-nsight in your project. Codex loads it when a task matches its description.

Can I use Replay Nsight in Cursor, Gemini CLI or GitHub Copilot?

Cursor, Gemini CLI, GitHub Copilot and OpenCode also load SKILL.md folders. With the skills CLI, run `npx skills add Luna5ama/Alpha-Piscium --skill replay-nsight -a cursor` (or -a gemini-cli, github-copilot or opencode for the others). To copy it by hand, put the folder in .cursor/skills/replay-nsight, .gemini/skills/replay-nsight, .github/skills/replay-nsight and .opencode/skills/replay-nsight in your project.

What does Replay Nsight need to run?

Going by SKILL.md and its folder, Replay Nsight needs PowerShell for the scripts in its folder. Our summary lists: PowerShell.

Does Replay Nsight access the network?

SKILL.md contains no URLs. Any network use would come from the scripts or tools the agent runs. This is read from the text; nothing was executed.

Is Replay Nsight safe to install?

Our automated static check of SKILL.md found no risky patterns, such as piping downloads into a shell, reading credential files or hidden Unicode. It is not a guarantee. The check reads SKILL.md only: the scripts in the folder are not scanned, so read them before running anything.

What licence does Replay Nsight use?

Replay Nsight is published under the GPL-3.0 licence (the repository's licence). It allows redistribution, so the full SKILL.md is shown on this page.

How many tokens does Replay Nsight use?

About 2.1k tokens (SKILL.md is roughly 8.5k characters). Agents keep only the skill's name and description in context until a task matches; then they load SKILL.md in full.

What are the alternatives to Replay Nsight?

Skills that share tags, products or a category with Replay Nsight: MCP Server Builder (anthropics/skills, 180k stars), MCP Server Builder (shareAI-lab/learn-claude-code, 78k stars), MCP Integration for Plugins (anthropics/claude-plugins-official, 38k stars) and Figma use_figma Plugin API Rules (warpdotdev/warp, 65k stars). The comparison table on this page puts their stars, adoption, token cost, safety result and licence side by side.

Who maintains Replay Nsight?

Luna5ama (a GitHub user) maintains it in Luna5ama/Alpha-Piscium, which has 156 GitHub stars. The repository holds 3 skills in this directory. The repository was last updated on October 5, 2026.

Source: Luna5ama/Alpha-Piscium on GitHub. Facts on this page come from the repository at the commit we read; the author's words are quoted as theirs.