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 well-engineered procedural skill: concrete commands at every step, real validation gates (duplicate check, parent verification, explicit user approval before creation), and clean structure. The small deductions are for a slightly wordy prerequisites fallback, one dubious GraphQL flag usage, and an inline-heavy layout for the label-mapping data.
Suggestions
Fix the issue-type lookup command: separate the flags as --jq '<filter>' --arg t "<type from template>" so the jq filter is not consumed as the --arg value.
Tighten the Prerequisites fallback paragraph to one sentence, e.g. "If gh auth fails, prefer available mcp__github__* tools; if neither works, retry once, then ask the user to check authentication."
Consider moving the component-label path mapping and the Zeebe engine/platform split into a references/ file, keeping SKILL.md as the procedure overview.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dominated by executable commands and two dense mapping tables that earn their tokens; it never explains concepts Claude already knows. A few sentences could be trimmed, e.g. "MCP tools use separate credentials, so they often keep working through a `gh` auth failure (stale token, sandbox restrictions, a transient rate limit), though they are not immune to failures of their own (their own rate limits, service-side restrictions)". | 4 / 5 |
Actionability | Nearly every step has a copy-paste-ready command (ls, gh search, gh issue view, gh api graphql, gh issue create with a mktemp body file). The one minor gap is the issue-type GraphQL lookup, where "--jq --arg t \"<type from template>\"" interleaves the --jq filter and --arg in a way that would not parse as written. | 4 / 5 |
Workflow Clarity | The seven steps are clearly sequenced with explicit checkpoints and feedback loops: duplicate search before drafting (Step 1.5), parent verification via gh issue view (Step 4), a hard approval gate ("Do NOT call `gh issue create` yet") with a revise-and-reask loop (Step 6), and auth-failure fallback handling in Prerequisites. | 5 / 5 |
Progressive Disclosure | As a single-file skill with no bundle files, all content is inline and organized under well-labeled sections, with the only external pointers (AGENTS.md permalink guidance, the grill-me skill) clearly signaled and one level deep. The ~240-line body is defensible since the mapping tables are needed on every run, though the component-label and Zeebe-layer mapping could arguably live in a reference file. | 4 / 5 |
Total | 17 / 20 Passed |