Content
82%Weight 40%Scale 1-5Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.
The body is a strong, execution-focused CLI reference: fully executable command examples for every subcommand, sensible sequencing guidance in Agent Notes, and clear routing between replay analysis and GPU Trace summarization. Weaknesses are minor — an inline version-specific caveat that belongs in a compatibility/deprecated note and no explicit output-verification checkpoint after captures.
Suggestions
Move the Nsight Graphics 2026.1.0 "Invalid file header" caveat into a labeled compatibility/deprecated section so time-sensitive version details don't clutter the main notes.
Add a validation checkpoint after capture commands (e.g. check that the output directory contains the expected .ngfx-capture/.ngfx-gputrace file before proceeding to summarize or analyze).
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is lean — copy-paste command examples with no filler and no explanation of concepts Claude already knows — but it embeds a time-sensitive version reference ("on Nsight Graphics 2026.1.0 they may report `Invalid file header`") outside any deprecated/old-patterns section, and a few Agent Notes (the enumeration of summary contents) could be tightened. | 4 / 5 |
Actionability | Every section provides fully executable, copy-paste-ready commands with real flags (--exe, --pid, --wait-frames, --auto-export, --summarize) and covers the common cases for each subcommand: doctor, launch, frame capture, GPU trace, replay analysis, and C++ capture. | 5 / 5 |
Workflow Clarity | Sequencing is explicit ("Prefer `doctor info` first to discover the available compatibility mode") with clear decision routing for input types and error-recovery notes (empty-table warnings, old-worktree reinstall), but there is no explicit post-capture verification step such as confirming output files were produced. | 4 / 5 |
Progressive Disclosure | No bundle files exist and none are needed; the body is well organized (Capabilities overview, per-command sections, Agent Notes). It falls short of a 5 because it is a ~95-line multi-command skill rather than the under-50-line single-task case, and all detail lives inline with no reference structure at all. | 4 / 5 |
Total | 17 / 20 Passed |