Content
85%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 content is highly actionable and operationally safe — executable commands, exact error strings, and validation guards for every failure mode of a batch fetch-and-render job. Its weaknesses are a dangling load-bearing reference (templates/changelog.md is absent from the bundle) and notable rule repetition across the Arguments, Examples, Anti-patterns, and Definition-of-done sections that inflates token cost without adding information.
Suggestions
Ship the referenced templates/changelog.md file (or inline the template verbatim in a Template section) — the skill instructs rendering 'verbatim' against a file that does not exist in the bundle, so the workflow dead-ends at the render step.
Deduplicate repeated rules: the scope-error message appears verbatim in three places and the Definition of done checklist restates the Arguments, Output rules, and Anti-patterns sections; consolidate to one authoritative statement per rule and reference it.
Move the ~100 lines of --audience=general and --scope=all rendering rules into a one-level-deep reference file (e.g. references/rendering-modes.md) so the core SKILL.md body stays a lean overview.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense with genuinely non-obvious, skill-specific rules (missing `mergedAt` in `gh search prs` JSON, inclusive `--closed` ranges, brand-name Title-casing, the 4-backtick outer fence) and assumes Claude's competence throughout. However, rules are repeated across sections — the scope-error string appears verbatim three times, and 'Definition of done' restates the Arguments, Output rules, and Anti-patterns sections — which keeps it below the fully lean anchor; it is above 'mostly efficient' because the padding is repetition of real rules, not explanation of concepts Claude already knows. | 4 / 5 |
Actionability | Guidance is copy-paste ready: a complete argument-parsing bash block with BSD/GNU date fallback, an exact `gh search prs` query with field list and repo-filter insertion, exact error-message strings, a literal truncation-warning line, and a first-match-wins bucketing lookup table. The only non-concrete area (Linear MCP tool names) is explicitly and justifiably deferred to runtime resolution ('resolve the list-issues and get-issue tools at runtime ... do not hard-code the namespace'), so no key details are missing. | 5 / 5 |
Workflow Clarity | A numbered six-step workflow with explicit validation checkpoints throughout: agent-side merged/closed partitioning by `state`, a truncation guard when results hit `LIMIT`, defined failure paths for missing/unauthenticated `gh` and unavailable Linear MCP (degrade with a notice, don't fail), a hard error for `--scope=current` outside a GitHub repo, an explicit empty-window state, and a closing checklist. As a batch read/render operation, every batch risk (truncation, empty results, source unavailability) has a defined response, so the batch-validation cap does not apply. | 5 / 5 |
Progressive Disclosure | In-file section structure is clean (Arguments, Workflow, Data sources, Feature grouping, Output rules, Template, Examples, Anti-patterns), but the body's single external reference — [`templates/changelog.md`](./templates/changelog.md), cited three times and required for rendering 'verbatim' — does not exist in the bundle (no templates/ directory), so the load-bearing reference is a dead path. Also, all ~365 lines of rules are inlined in one file with the ~100 lines of scope/audience rendering rules as candidates for a separate reference file. | 3 / 5 |
Total | 17 / 20 Passed |