Content
100%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.
An exemplary orchestration skill body: fully templated commands with per-case flag variants, an explicitly sequenced eleven-step workflow with dry-run/confirm gates around all outward-facing Jira writes, and clean one-level-deep references into a real bundle. The only trivial imperfection is the ellipsis in the `new-plan.ts` invocation.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Lean throughout: no concept explanations Claude already knows, no filler — e.g. "Run one investigation per report. Never change product code." and "Don't guess a verdict to finish the run." are one-line rules. The opening two-line framing ("why did it break now, and will it keep breaking") states the skill's mental model at minimal token cost; every token earns its place. | 5 / 5 |
Actionability | Nearly every step is a copy-paste-ready command with full argument templates and flag variants: `$B/bug-intake.ts --issue <N|url> | --jira <KEY|url> | --file <path> | --text "<report>"`, `$H/analyze.ts --run $RUN --focus <every path file> --label "<report title>"`, and `$S/bug-verdict.ts` with per-verdict flag lists (one-off: `--fix`, pattern: `--plan`, known: `--ticket`, needs-info: `--ask`). The common cases are fully covered; only `$P/new-plan.ts --run $RUN ...` elides arguments. | 5 / 5 |
Workflow Clarity | Eleven clearly sequenced steps with explicit validation checkpoints: preflight "Stop on any nonzero exit", dry-run discipline for outward-facing writes ("first run the same command without `--apply`, show the user the dry run, and apply only after they confirm"), a re-run feedback loop ("Re-run `analyze` as the trace grows"), and a completion check ("Exit 1 means no verdict yet"). The Jira-writing operations all pass through the dry-run/confirm gate. | 5 / 5 |
Progressive Disclosure | The body is a clean orchestration overview: decision criteria are split into `references/bug-verdicts.md` (verified to exist), detailed investigation method into `.agents/skills/bug-trace/references/bug-investigation.md`, and scripts are addressed via the path-variable block — all one hop from SKILL.md, clearly signaled, and every referenced bundle path (`references/bug-verdicts.md`, `scripts/bug-verdict.ts`, `scripts/summarize.ts`) resolves to a real file. | 5 / 5 |
Total | 20 / 20 Passed |