Content
96%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 body is an exemplary operational skill: lean, fully executable bash with error handling, a clearly sequenced workflow with validation checkpoints, guarded state restoration, and explicit exit criteria. The only notable weakness is that all content lives inline in SKILL.md with no reference/script split, which is acceptable for a single linear workflow but leaves some reusable blocks inlined.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is lean and imperative throughout — no explanations of concepts Claude already knows, no padding. Every line carries project-specific knowledge (e.g., "Running workers hold stale code", "A success here never refutes a reporter failure on a different account", the hard-flags list). This matches anchor 5 (lean, assumes Claude's competence, every token earns its place); there are no minor over-explanations that would justify a 4. | 5 / 5 |
Actionability | The skill is fully executable and copy-paste ready: concrete gh/jq commands, guarded checkout with a cleanup trap, port probing, a server readiness loop with failure output ("Server died during startup. Tail of ..."), the full MCP initialize/initialized/tools/call handshake, and a diff step. Common and failure cases are both covered, matching anchor 5. It is not a 4 because there are no gaps — even error paths include what to print and when to exit. | 5 / 5 |
Workflow Clarity | The four numbered sections form a clear sequence with explicit validation checkpoints at every risky step: input parsing fails fast ("Invalid input: '$ARGUMENTS'"), a dirty-tree check aborts checkout, baseline files are checked with [ ! -s ] before any live replay, server readiness and Mcp-Session-Id are verified, and protocol errors are classified as execution limits. There are feedback loops ("ask before checkout", "If either file is missing... Offer the evidence-and-code path"), per-step exit criteria ("This step is complete when..."), a defined outcome taxonomy, and a Non-negotiables checklist. This matches anchor 5. | 5 / 5 |
Progressive Disclosure | There are no bundle files (no references/, scripts/, or assets/), and the body is well-sectioned with clear headers and each block placed where the workflow needs it. However, the skill is ~160 lines, and heavyweight inline blocks — the ~30-line MCP startup/handshake sequence and the audit hard-flags — could plausibly live in shared scripts/reference files for reuse across the skill family. That minor organization gap matches anchor 4 (good structure, most content appropriately placed) rather than anchor 5 (well-signaled one-level-deep references with appropriate split). | 4 / 5 |
Total | 19 / 20 Passed |