Content
73%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 strongly actionable, well-sequenced contract-style skill body: concrete CLI commands and formats, phased workflow with explicit readback verification, and thorough error-recovery guidance keyed to exit codes. Its weaknesses are repeated consent/convention statements that inflate token cost and reference paths that escape the skill's own bundle.
Suggestions
Consolidate the thrice-repeated consent rule into a single authoritative statement (e.g., in the Enablement section) and reference it from the Contract and When-it-fails lists instead of restating it.
Trim the harness-convention paragraph to its operative rule (convention plus optional Stop-hook backstop, not a mechanical guarantee) or move the nuance into a reference file.
Add one concrete example of a `put_page`/`add_link` MCP invocation so every named tool has executable usage, and either bundle the conventions docs under `references/` or drop the dangling cross-bundle paths.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense and assumes agent competence, but the consent rule ("an API key, silence or a first-fire announcement is not consent") is repeated across the Contract, Enablement, and When-it-fails sections, and the harness-convention paragraph restates the same convention-vs-guarantee point several times within itself. Consolidating the repetitions would tighten the document noticeably. | 3 / 5 |
Actionability | Concrete, executable guidance throughout: `gbrain search "name"`, `gbrain timeline-add <slug> <date> "<summary>"`, explicit slug targets (`originals/{slug}`, `concepts/{slug}`, `ideas/{slug}`), a ready-to-use back-link format template, and signal-log line examples. The minor gap is that MCP tool usage (`put_page`, `add_link`) is named but never shown with invocation syntax or examples. | 4 / 5 |
Workflow Clarity | Phases are clearly sequenced (idea detection PRIMARY, entity detection SECONDARY, logging always), the Output Format section mandates a readback verification of captured content, and the "When it fails" section gives explicit per-error recovery loops (e.g., `write_pending` exit 10 → poll the receipt or re-read and merge next turn) plus a pointer to the operator protocol, satisfying validation and feedback-loop requirements for a batch-writing skill. | 5 / 5 |
Progressive Disclosure | A well-sectioned single file with no bundle files to split, and clearly signaled one-level references to conventions docs (`skills/conventions/quality.md`, `../conventions/agent-operator-protocol.md`). However, those references point outside this skill's bundle (no `references/` directory exists), and the long inline harness-convention paragraph is material that arguably belongs in a reference file. | 4 / 5 |
Total | 16 / 20 Passed |