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 tight, highly actionable instruction skill: executable gh/jq commands, a concrete rating rubric, and clearly sequenced modes with a built-in confirm-or-mark-unconfirmed safety design. The gaps are minor — a missing post-apply verification step in the batch backfill pass and a few rhetorical passages that could be cut.
Suggestions
Add a verification step to Mode B's batch pass (e.g. re-run the listing filter after applying to confirm no issues remain unrated, or spot-check a sample of applied ratings) so the batch operation has an outcome check.
Trim the rhetorical explanations in 'Why the timing is the whole point' — e.g. the 'talked into agreeing with the spend' imagery and the motivated-reasoning aside — down to the bare rules they illustrate.
Signal the trailing 'The labels' and 'What consumes this' sections as reference material (or fold them into a short 'Reference' heading) so the operational core reads as Modes A and B first.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Mostly lean and project-specific — the timing rationale, rubric table, and gh commands earn their tokens. Not 5: a few rhetorical passages could be trimmed, e.g. "the number gets talked into agreeing with the spend" and the aside explaining that self-rating is "motivated reasoning", a concept Claude already knows; not 3: no padded sections or teaching of background concepts. | 4 / 5 |
Actionability | Fully executable throughout: copy-paste gh commands with exact label and repo arguments ("gh issue edit 412 --repo englishstreetventures/osn --add-label \"complexity:3,complexity:unconfirmed\""), a jq listing filter for unrated issues, and a concrete Fibonacci rubric with repository-shaped examples per row. Not 4: the commands and decision criteria cover the common cases with no gaps. | 5 / 5 |
Workflow Clarity | Mode A is a numbered sequence with an explicit human checkpoint (owner confirm or amend, with neighbouring values offered) and a defined unattended fallback; Mode B pre-filters to unrated issues and carries the unconfirmed-marker safety design. Not 5: the batch backfill pass has no post-apply verification (e.g. re-checking that labels landed or spot-checking a sample); not 3: checkpoints are present and explicit — the jq select gates the batch and the unconfirmed label makes unattended output separable. | 4 / 5 |
Progressive Disclosure | A single ~110-line file with clean section headers (rubric, Mode A, Mode B, labels, consumers) and no bundle files; everything is load-bearing and easy to navigate. Not 5: it exceeds the under-50-line simple-skill case and the trailing labels/consumer detail could be more clearly signaled as reference material; not 3: no content that clearly belongs in a separate file is inlined and nothing is buried. | 4 / 5 |
Total | 17 / 20 Passed |