Content
92%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-sequenced, executable runbook with concrete commands, a worked example, and strong validation/feedback loops for the outward-facing review-posting operation. It is concise without sacrificing actionability and is cleanly sectioned.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Largely lean and procedural with no padding about concepts Claude already knows; a few framing sentences (e.g. 'The whole job is…') and the rationale asides could be trimmed slightly, keeping it just under the top anchor. | 4 / 5 |
Actionability | Provides fully executable, copy-paste-ready commands (git clone/fetch/checkout, curl GET/POST) plus a worked JSON review example with concrete line numbers — directly actionable guidance covering the common cases. | 5 / 5 |
Workflow Clarity | An 8-step sequenced workflow with explicit validation checkpoints: head-sha mismatch is noted, clone/diff failures block the card, a 422-recovery feedback loop with up-to-3 retries, and a hard gate ('You MUST NOT call kanban_complete unless this check passes'). | 5 / 5 |
Progressive Disclosure | No bundle files exist, so it is scored against its own structure; the body is organized into clearly numbered sections and language-reviewer detail is delegated one level deep to $HERMES_HOME/skills/<name>/SKILL.md with no nested references. | 5 / 5 |
Total | 19 / 20 Passed |