Content
81%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 tight, high-quality instruction-only skill: fully executable commands, an enumerated failure matrix, and correct offloading of voice skeletons to a real bundle reference. The main improvement levers are consolidating the repeatedly restated write-guard constraints and specifying the readout slug derivation.
Suggestions
Consolidate the hot.md / _readouts/ / index.md do-not-touch rules into a single 'Write Guards' section instead of restating them across Retrieval, Persistence, Logging, and Safe Failure.
Specify the deterministic slug rule (e.g., lowercase topic, spaces to hyphens, stripped of path characters) so the _readouts/<slug>.md filename is unambiguous.
Clarify where llm-wiki/SKILL.md, WRITING.md, and the Config Resolution Protocol live, or inline the minimal needed details, so the skill is self-contained when the sibling bundle is not present.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense and prescriptive with no concept-explanation padding, but the hot.md/_readouts_/index.md do-not-touch constraints are restated across the Retrieval, Persistence, Logging, and Safe Failure sections and could be consolidated. Not 3 because every other token is load-bearing; not 5 because the repetition is a real trimming opportunity. | 4 / 5 |
Actionability | Concrete throughout: exact command syntax, a copy-paste-ready log command with all flags, explicit exclusion lists, marker syntax (^[inferred], ^[ambiguous]), and saved-readout frontmatter fields. Minor gaps keep it from 5: the <slug> is described as deterministic and filesystem-safe but never specified, and external dependencies (llm-wiki/SKILL.md, Config Resolution Protocol) cannot be resolved from this bundle. | 4 / 5 |
Workflow Clarity | The flow (contract → retrieval → ledger → draft → persist → log) is clearly sequenced, and the Safe Failure Behavior section enumerates explicit error-recovery paths for no-match, weak evidence, QMD fallback, write failure, and log failure — feedback loops and validation checkpoints are fully present. | 5 / 5 |
Progressive Disclosure | Voice skeletons are appropriately split into references/voices.md, which exists in the bundle and is clearly signaled one level deep (verified). Not 5 because several referenced resources (llm-wiki/SKILL.md, WRITING.md, AGENTS.md, .manifest.json) live outside this bundle and are not navigable or verifiable from it. | 4 / 5 |
Total | 17 / 20 Passed |