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.
The body delivers exceptionally actionable, well-sequenced operational guidance with real validation checkpoints and error-recovery loops — the learning-loop protocol in particular is unambiguous and executable. Its weaknesses are size-related: a ~420-line monolithic file with no progressive disclosure into bundle files, and a few padded or duplicated passages. Splitting reference material out and trimming the meta-commentary would raise both conciseness and structure without losing any operational value.
Suggestions
Move the stable reference material — the recall response envelope JSON schema, exit-code table, CSV-formula-injection contract, MCP HTTP/origin configuration, and path-resolution rules — into a references/ file (e.g., references/cli-contract.md) and keep SKILL.md as the operational core with clearly signaled one-level-deep links.
Delete the duplicated read-only prohibition (keep the 'When Not to Use This CLI' section, drop the repeated list under Agent Mode) and cut the generator meta-commentary about the sync-walker fixture and EXAMPLE-*/ALPHA/BRAVO identifiers, which an operator never needs.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Most sections are dense, command-first, and assume Claude's competence, but there is removable fat: the 'Golden fixture... mirrors the sync-walker fixture' paragraph is generator meta-commentary irrelevant to operating the CLI, and the read-only prohibition is stated twice (the 'When Not to Use This CLI' section and again as 'Read-only — do not use this CLI for create, update, delete...' under Agent Mode). It is not 4 because those sections can be cut without losing guidance; not 2 because most of the body is tight and operational. | 3 / 5 |
Actionability | Guidance is fully executable and copy-paste ready throughout: the npx install command, `echo "$TOKEN" | learn-loop-example-pp-cli auth set-token`, the file-then-$QUERY recall/teach/playbook-amend invocations, the which/auth/profile/feedback command sets, and the exit-code table. Specific examples cover the common cases, matching the top anchor. | 5 / 5 |
Workflow Clarity | The learning loop is sequenced as six explicit steps (recall → decision tree → warnings → teach → playbooks → amend) with concrete validation checkpoints ('You must verify the CLI is installed before invoking any command', 'check that the printed payload matches the behavior you verified', 'Do not proceed with skill commands until verification succeeds') and error-recovery feedback loops (confirm/reject, graceful degradation for older binaries, empty-store short-circuit). No destructive or batch operation lacks validation — even `local-data purge` previews and requires `--yes`. | 5 / 5 |
Progressive Disclosure | Section headers are clear and in-file navigation is good, but the skill is a single ~28KB monolith with no bundle files at all (no references/, scripts/, or assets/ exist), while content that clearly belongs in separate files is inlined: the full recall-envelope JSON schema, exit-code and CSV-injection contract details, MCP HTTP transport configuration, and path-resolution internals. It is above anchor 2 because the structure is genuinely good rather than minimal, but below 4 because the split is wholly absent, not a minor gap. | 3 / 5 |
Total | 16 / 20 Passed |