Content
96%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 excellent procedural skill body: fully executable commands, a hard approval gate, explicit validation with fix-and-retry loops, and disciplined token efficiency. The only structural weakness is that the two key referenced prompt files are not present in the bundle, leaving the disclosure structure partially unverifiable.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is lean and imperative throughout — exact commands, filenames and line-level rules with no explanation of concepts Claude already knows; the few justifications (the shell-injection warning for typed titles, why the draft lines must not reach GitHub) state non-obvious operational facts and earn their tokens. | 5 / 5 |
Actionability | Nearly everything is copy-paste executable: exact git/gh commands, a runnable python heredoc CI check, and a complete awk/shell block for extracting the title and body; steps 3 and 4 delegate to named prompt files with concrete pointers rather than vague direction. | 5 / 5 |
Workflow Clarity | The 12 steps are clearly sequenced with explicit validation checkpoints and feedback loops: step 8's CI check with 'Fix tmp-PR-content.md until the check passes', re-validation after developer edits in step 10, a hard approval gate ('Do not continue until the developer writes approve'), and safeguards before overwriting an open PR body. | 5 / 5 |
Progressive Disclosure | Sections (Inputs, Steps, Rules) are well organized and references are one level deep and clearly signaled ('Obey pr-what-prompt.md in this folder'), but the actual bundle contains only SKILL.md — the referenced pr-what-prompt.md and release-note-prompt.md files are absent, so the navigation targets cannot be verified. | 4 / 5 |
Total | 19 / 20 Passed |