Content
78%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 delivers concrete, executable, well-sequenced guidance for all four modes with real git commands and precise formatting rules. Its main defect is bundle integrity: every section defers to references/conventional-commits.md, which does not exist, leaving the promised detail unreachable; a smaller gap is the missing validate-after-generate feedback loop.
Suggestions
Create the missing references/conventional-commits.md with the sections the body cites (Type Classification, Scope Rules, Message Structure, Body and Footer Rules, Atomic Commits, Validation Rules), or inline the essential rules and drop the reference lines.
Add a feedback loop to generate mode: after producing the message, run validate mode against it and fix any violations before presenting the final output.
Trim known-concept explanations — the imperative-mood walkthrough and the per-type glosses — to just the allowed type list and a single style example.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Largely efficient — it assumes git knowledge and gives terse numbered steps with inline commands rather than prose. Minor over-explanation could be trimmed: the imperative-mood walkthrough ("'add', 'fix', 'remove' — not 'added', 'fixes'") and the one-line glosses on standard conventional-commit types restate concepts Claude already knows, though the type list doubles as the allowed-list referenced by validate mode. Not 5 because of those trimmable passages; not 3 because padding is minor and localized. | 4 / 5 |
Actionability | Fully executable guidance: exact commands (`git diff --staged`, `git diff HEAD`, `git status`, `git add <files>`, `git diff --staged --stat`, `git log -1 --format=%B`), a concrete subject-line template `<type>(<scope>): <imperative verb> <what and why>` with hard 72-char limits, a good-vs-bad example, and the full message structure in a copy-paste-ready block. Per the rubric's code-vs-instruction note, an instruction-only skill with this level of specific guidance earns the top anchor. | 5 / 5 |
Workflow Clarity | Each mode has a clear numbered sequence with some explicit checkpoints — analyze's conditional fallback ("if empty, run `git diff HEAD`"), stage's confirmation via `git diff --staged --stat` and its ask-which-group prompt, and validate's PASS/violations report with corrected lines. Not 5 because generate mode ends at output without a feedback loop (e.g., "run validate mode on the generated message"), a minor validation gap that fits anchor 4. | 4 / 5 |
Progressive Disclosure | The body is well-structured into four mode sections, each ending with a clearly signaled one-level-deep reference line ("Reference: `references/conventional-commits.md` → Type Classification, Scope Rules"), but scoring against the actual bundle shows no `references/` directory exists — all four references point to a missing file, so the deferred detail (Validation Rules, Scope Rules, Atomic Commits) is inaccessible. Not 4 because the disclosure structure is only partially realized; not 2 because signaling and in-body organization are good and content is not wrongly inlined. | 3 / 5 |
Total | 16 / 20 Passed |