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.
A tightly authored, highly actionable workflow document: concrete commands in every phase, explicit exit conditions and a CI failure feedback loop, consent boundaries, and a well-timed reference table that keeps the body to a lean overview. The only weakness is external: the plugin-root paths (assets/writing.md, scripts/prose-lint.py) that the workflow depends on at three points are not resolvable within the skill's own bundle.
Suggestions
Inline the essential writing rules from ../../assets/writing.md (or copy the file into the skill's references/) so the commit/PR text guidance is reachable from the skill bundle itself, since it is required at both phase 2 and phase 5.
State the expected location or fallback for scripts/prose-lint.py (e.g., 'if absent, skip linting and say so') so the mandatory lint steps in phases 2 and 5 cannot dead-end when the plugin script is not installed.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is lean and prescriptive throughout: it assumes Claude knows what PRs, CI, and Conventional Commits are and never explains them, spending its ~90 lines entirely on rules, commands, and exit conditions (e.g., 'Never declare success with checks in flight.'). The short intro and forge note ('use the forge's equivalent, such as `glab`') each earn their tokens; this matches the lean-and-efficient top anchor rather than the minor-trimming-warranted anchor below it. | 5 / 5 |
Actionability | Nearly every phase gives copy-paste-ready commands: `touch "$(git rev-parse --git-dir)/swd-submit"`, `git log @{u}..HEAD`, `gh pr view --json number,title,body,state,isDraft,baseRefName`, `git push -u origin <branch>`, `gh pr create --draft`, `gh pr edit`, `gh pr checks`, and `prose-lint.py --commit`. Even branch naming is concrete ('`<author>-<slug>`: the git user's first name in lowercase'), and detail is correctly deferred to reference files at the phase where it is needed — matching the fully-executable top anchor. | 5 / 5 |
Workflow Clarity | Phases 0-6 are strictly sequenced with an explicit sequencing rule ('Each phase meets its exit condition before the next one starts'), per-phase exit conditions, a re-entry rule for re-invocation, and an explicit CI feedback loop: 'On failure, read the failing logs, find the root cause, fix it, and re-enter at phase 3.' A 'Definition of done' checklist closes it out — matching the top anchor for explicit validation steps and error-recovery loops, which matters given the destructive/batch-adjacent operations (pushing, staging). | 5 / 5 |
Progressive Disclosure | The references table with a 'Read at' timing column is exemplary signaling — each of the four references/ files exists in the bundle, is one level deep, and is loaded at a named phase. However, two frequently referenced paths, `../../assets/writing.md` (referenced in phases 2 and 5 and in the reference files) and `<plugin root>/scripts/prose-lint.py`, live outside the skill's directory and are absent from the provided bundle, so a reader following them from the bundle hits dead ends. This fits 'good structure... references mostly clear; minor organization gaps' rather than the flawless top anchor. | 4 / 5 |
Total | 19 / 20 Passed |