Content
75%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 content is well-organized and actionable with lean prose and a clear goal-driven workflow, scoring consistently at 4 across dimensions. The main gap is that several referenced bundle files (scripts, references, hooks) are advertised but absent, which weakens progressive disclosure.
Suggestions
Ship the referenced bundle files (references/karpathy-principles.md, references/anti-patterns.md, references/enforcement-patterns.md, scripts/*.py, hooks/karpathy-gate.sh) so the documented navigation resolves, or remove references to files that do not exist.
Inline at least one short executable example (e.g. a sample complexity_checker.py invocation and its output) so the body is copy-paste ready without relying on external --help.
Add an explicit validate->fix->retry loop to the main /karpathy-check review workflow rather than only to the per-task plan template.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is lean—bullet principles, compact tables, minimal prose—and avoids explaining concepts Claude already knows, but the multi-line Karpathy quote block and the supplementary cross-tool/related-skills sections add length that could be trimmed. Not 5 because some sections are supplementary rather than strictly necessary; not 3 because there is no real padding. | 4 / 5 |
Actionability | Provides concrete commands (/karpathy-check, --help), a tools table stating what each detects, the hook wiring path, and a plan template with verify checks, but contains no inline copy-paste code sample since the scripts live externally. Not 5 because it relies on external --help rather than inline executable examples; not 3 because the guidance is mostly executable and concrete. | 4 / 5 |
Workflow Clarity | The goal-driven section gives an "Instead of -> Transform to" table and a numbered plan template with a verify checkpoint per step, but the main review flow (/karpathy-check -> sub-agent -> tools) is described at a high level without an explicit validate->fix->retry error-recovery loop. Not 5 because the primary workflow lacks explicit recovery checkpoints; not 3 because checkpoints are present via the plan template. | 4 / 5 |
Progressive Disclosure | The body is well-structured with a clearly signaled References section and a scripts table pointing one level deep to references/*.md, scripts/*.py, and hooks/*.sh, but the referenced bundle files (e.g. references/anti-patterns.md, complexity_checker.py) do not exist on disk, so the claimed navigation does not resolve. Not 5 because the promised references are missing; not 3 because the written structure is genuinely well-organized rather than inline-dumped. | 4 / 5 |
Total | 16 / 20 Passed |