Content
100%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 exemplary: executable commands with flags and exit codes, hard rules stated as rules rather than tutorials, and a verified one-level-deep bundle split. It teaches only what is specific to these three custom scripts and defers everything else to well-labeled reference files.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The ~66-line body is dense with script-specific facts Claude cannot infer (which checks block vs warn, CLDR 49 embedded, exit codes, growth-rate padding) and explains no general i18n concepts. Nothing reads as padding, matching the 'every token earns its place' anchor. | 5 / 5 |
Actionability | The Workflow section gives three copy-paste commands with real paths and flags (--source-locale en, --bidi, -o), the Scripts table documents exit codes and universal flags, and remediation is concrete (--warn-only, i18n-ignore per line, .i18n-allow for brand names). Fully executable guidance covering the common cases. | 5 / 5 |
Workflow Clarity | A clearly numbered 1-2-3 sequence where the scripts are themselves validators with documented blocking exit codes, step 3 is an explicit verification pass ("click through the app in it" with interpretation rules like "Plain English on screen is a hardcoded string"), and finding-remediation loops (ignore, allowlist, mark gaps for a translator) are stated. The checks are read-only, so the destructive/batch cap does not apply. | 5 / 5 |
Progressive Disclosure | The body is a lean overview; all three references, both assets, and all three scripts exist on disk and are one level deep (no nested references inside them), each listed with a per-file scope description and source count. Well-signaled, easy navigation matching the top anchor. | 5 / 5 |
Total | 20 / 20 Passed |