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.
The body is a well-engineered operational playbook: fully concrete commands, an explicit five-step sequence with validation checkpoints and error-recovery loops, and accurate condition-scoped pointers to a real three-file reference bundle. Its only weakness is mild phrasal looseness in a handful of steps that could be tightened without losing meaning.
Suggestions
Tighten Step 2.2 ('Keep `name` and `version` present at the top level; treat them as required contract fields') to a single clause, and trim the directory enumeration in Step 1.1 to the few paths that actually change the decision.
Compress Step 4.4's explanation of what `apm install` already deploys into a one-line contrast with `apm compile` (e.g. 'compile only generates instruction files like AGENTS.md/CLAUDE.md; install already deploys everything else').
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense and command-focused with no explanations of concepts Claude already knows — every step names a concrete CLI action. Minor trimmable instances remain, e.g. 'Keep `name` and `version` present at the top level; treat them as required contract fields' (the second clause restates the first) and the long directory-enumeration list in Step 1.1. Not 5 because a few phrases are restated or spelled out beyond what a competent reader needs; not 3 because there is no genuinely unnecessary explanation or padding. | 4 / 5 |
Actionability | Guidance is fully executable throughout: exact commands (`apm install --dry-run`, `apm compile --validate`, `apm preview <script> -p key=value`, `apm mcp show <server>`) and a concrete worked example (`apm install webmaxru/agent-skills/skills/webmcp`) that covers the common cases. Not 4 because the commands are copy-paste ready and include flags, arguments, and a realistic package-path example with no gaps in the core flows. | 5 / 5 |
Workflow Clarity | A clearly sequenced five-step procedure with explicit validation checkpoints and feedback loops for risky/batch operations: dry-run previews before changes, 'verify the lockfile `resolved_commit` actually changed' after updates, 'verify the resolved state with `apm deps list`/`tree`/`info`', `--watch` only after the one-shot validation path is clean, and `apm pack` only after a successful install — plus a dedicated Error Handling section with fix-then-retry loops. Not 4 because validation is explicit rather than implicit and error recovery is a first-class section, matching the top anchor. | 5 / 5 |
Progressive Disclosure | The SKILL.md body is a lean overview that defers detail to real, one-level-deep bundle files, each clearly signaled with a conditional trigger: references/manifest-and-lockfile.md, references/command-workflows.md, references/troubleshooting.md, and assets/apm.yml.template — all verified present on disk, and named troubleshooting sections ('Stale packages after update', 'Self-referencing dependencies') exist in the referenced file. Not 4 because every reference is condition-scoped, accurate, and exactly one level deep with no inlined content that belongs in a separate file. | 5 / 5 |
Total | 19 / 20 Passed |