Content
63%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 with genuine validation checkpoints, especially in the learning-loop protocol. Its weaknesses are redundancy and monolithic structure: ~400 lines with no reference files, duplicated statements, and a misplaced TLS paragraph.
Suggestions
Move the learning-loop protocol (Steps 1-6, the recall response envelope, and the warnings catalog) to a references/ file, keeping a short protocol summary in SKILL.md that points to it.
De-duplicate the read-only prohibition and the credentials.toml location, and relocate the TLS/CA-bundle paragraph to a relevant section instead of "When Not to Use This CLI".
Remove the repeated fixture sentence from the body and add a minimal playbook-file JSON example so the `--playbook-file` path is fully actionable.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense with operational specifics rather than explanations of known concepts, but contains clear redundancy: "Purpose-built fixture for rich auth env-var model coverage." is repeated verbatim, the read-only prohibition appears in both "When Not to Use This CLI" and the "Read-only" bullet, the credentials.toml location is stated in two sections, and a TLS paragraph is misplaced under "When Not to Use". Mostly efficient but needs tightening. | 3 / 5 |
Actionability | Commands are largely copy-paste ready: the npx install line, `--version` verification, recall/teach/playbook amend invocations with argv-safe patterns, an exit-code table, and the `--deliver` sink table. Minor gaps: the `--playbook-file ~/playbooks/<shape>.json` shape is described but never shown with a concrete example. | 4 / 5 |
Workflow Clarity | The learning loop (Steps 1-6) is explicitly sequenced with a decision tree, and validation is present throughout: mandatory install verification ("Do not proceed with skill commands until verification succeeds"), `learnings confirm` printing the payload to check before materializing, graceful degradation for older binaries, and `doctor` checks. Falls short of 5 because misplaced and duplicated sections (TLS paragraph under "When Not to Use") leave minor gaps in the top-level flow. | 4 / 5 |
Progressive Disclosure | No bundle files or reference files exist; all ~400 lines are inlined in a single SKILL.md. The learning-loop protocol (~180 lines), MCP server setup, and output-delivery details clearly belong in separate reference files. Section headers provide reasonable structure, so this lands at "some structure but content that should be separate is inline" rather than minimal structure. | 3 / 5 |
Total | 14 / 20 Passed |