Content
82%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 high-quality, opinionated skill body: nearly every paragraph carries project-specific rules Claude could not infer, backed by executable code, runnable verification greps, and concrete anti-patterns with real consequences. The only costs are a few anecdotal passages that could be tightened and a 240-line monolith that could offload some deep-dive material to reference files.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense with project-specific, non-inferable rules (level semantics, anti-patterns, DI wiring) and never explains generic concepts Claude already knows. Minor trimmable padding keeps it from a 5: the Whispering incident narrative ("Whispering did exactly this and grew a parallel log..."), and the jsonlFileSink removal history ("This used to be jsonlFileSink; that primitive was removed because...") are longer than the rule they justify. Clearly above the 3 anchor, which flags whole sections of unnecessary explanation. | 4 / 5 |
Actionability | Quotes: the quickstart import block, the level table with exact signatures, the trySync call-site example, `grep -rn "consoleSink({" src/` and `grep -rn "log\.(warn|error)(new Error" src/` as runnable pre-checks, and the memorySink test snippet with assertions. Code is copy-paste ready and the examples cover the common cases (quickstart, branching, sinks, DI wiring), matching the 5 anchor. | 5 / 5 |
Workflow Clarity | For a guidance/reference skill, the decision flow is unambiguous: pick a level from the table, use the dominant branch-on-Result shape, avoid the two named anti-patterns (with grep verification commands as checkpoints), wire sinks via DI. Not a 5 because there is no explicit multi-step sequence with feedback loops, and no validation step for the sink-serialization guidance; comfortably above a 3 since the grep pre-checks and the level table serve as concrete decision checkpoints. | 4 / 5 |
Progressive Disclosure | Well-organized single-file skill with clear section headers, a quickstart up front, and clearly signaled cross-skill references ("error-handling skill: ... the [error variants reference](../error-handling/references/error-variants.md)", "rust-errors skill: full tracing ↔ Logger mapping") that are one level deep. Not a 5 because at ~240 lines with no bundle files, deep-dive material like the anti-pattern incident detail and the variant-naming rules could arguably live in a references/ file; well above a 3 since the inline content is curated guidance rather than an API-reference dump and navigation is easy. | 4 / 5 |
Total | 17 / 20 Passed |