Content
88%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.
An exceptionally operational skill: every step is an exact command with explicit refusal, crash-recovery, and QA-verification checkpoints, and it correctly delegates all decision logic to a tested script rather than restating rules. The minor costs are slight redundancy in the public-record warnings and a single-file body whose per-mode rule lists make it a touch denser than ideal.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense and imperative with zero explanation of concepts Claude already knows; the only padding is the deliberate repetition of the public-record rule ("Never include private logs, run artifacts, credentials..." vs "never include private logs, run artifacts or screenshots' file paths") and a few overloaded sentences that could be tightened. Not 5: that repetition and sentence density mean not every token earns its place; not 3: there is no unnecessary explanation anywhere. | 4 / 5 |
Actionability | Fully executable throughout: exact commands with the required env var ("CARNIVAL_APP_WRITER_HELD=1 node scripts/native/native.mjs backlog <file>"), the exact batch JSON shape, the exact gh api pagination command, and concrete capture invocations — copy-paste ready with the common cases covered. | 5 / 5 |
Workflow Clarity | Three clearly sequenced MODE sections with explicit validation and feedback loops for batch operations: the script "refuses duplicates by title or scope", "When it reports `partial` with errors, fix the candidates it names and run it once more", crash-rerun guidance ("After a crash, rerun the same file"), an uncertain-state single rerun rule, and browser-QA verification gates before claiming a PR shipped. | 5 / 5 |
Progressive Disclosure | A single well-sectioned file with no nested references; external materials (docs/visual-baseline/README.md, WORKSHOP.md, game READMEs) are signaled inline at the point of use, and decision logic is correctly delegated to the tested script. Not 5: the body is ~90 lines — beyond the under-50-line simple-skill exception — and per-mode rule lists (e.g. the MODE=backlog candidate rules) run long enough that modest splitting or tightening would improve navigation. | 4 / 5 |
Total | 18 / 20 Passed |