Content
77%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 highly actionable, well-sequenced skill with executable recipes and an unusually strong validation culture (Delivery Gates, known-issue workarounds). Its weaknesses are token efficiency — the same trap knowledge is restated several times — and the total absence of progressive disclosure: 45KB of reference-grade material is inlined in a single SKILL.md with no bundle files.
Suggestions
Split reference-grade material out of SKILL.md into bundle files: move the 19-row Known Issues table to references/known-issues.md, the Path B raw-set recipes and XML fragments to references/raw-set.md, and the MERGEFIELD pattern table to references/fields.md, leaving a one-line summary plus pointer per topic in the body.
Deduplicate the protection=forms semantics — currently stated in the Mental Model, the 'protection=forms is the LAST structural command' section, the Protection modes KEY, and the lock×protection interaction section — into one canonical section and cross-reference it elsewhere.
Move the inline Python XML-extraction snippet (block-level lock recipe) into scripts/ (e.g. scripts/extract_paragraph.py), invoking it with one line instead of a 10-line heredoc.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense with genuinely non-obvious CLI knowledge (shell-escape layers, K1–K19 issues, batch traps), but the same facts are repeated: protection=forms semantics are restated in the Mental Model, the "LAST structural command" section, the Protection modes KEY, and the lock×protection interaction table; the checkbox-is-formfield-not-SDT rule appears ~5 times; the file is 45KB. Matches "mostly efficient but includes some unnecessary explanation or could be tightened"; not 4 because the repetition is substantial, not minor trimmings. | 3 / 5 |
Actionability | Fully copy-paste-ready: exact bash commands, xpaths, namespace-carrying XML fragments, readback expectations ("expect: type=dropdown items=Engineering,Finance,HR"), decision tables per need, and a six-gate executable Delivery Gate script covering validation, leak detection, field inventory, and protection checks. Matches the fully-executable anchor with common cases covered. | 5 / 5 |
Workflow Clarity | An explicit build order ("create+open → metadata → structure → … per-field lock → protection=forms LAST → close") plus the Delivery Gate section gives explicit validation steps with REJECT/exit-1 feedback loops and symptom→fix tables for raw-set failures and WARNING: UNSUPPORTED. Matches the clear-sequence-with-feedback-loops anchor. | 5 / 5 |
Progressive Disclosure | No bundle files exist (references/, scripts/, assets/ are all absent), so all 45KB lives inline in SKILL.md: a 19-row Known Issues table, three Path B raw-set recipes, MERGEFIELD pattern tables, and an inline Python XML-extraction snippet — reference material that belongs in separate files. Section headers, decision tables, and a Quick Start keep it navigable, matching "some structure but … content that should be separate is inline"; not 2 because structure and navigation are far better than the header-less inlined-API-reference example, not 4 because the monolithic 45KB single file is the core anti-pattern this dimension exists to catch. | 3 / 5 |
Total | 16 / 20 Passed |