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 strong, highly actionable body: dual-language copy-paste code for all three invocation patterns, a genuine traps section with error-recovery feedback loops, and a production checklist. The main costs are token duplication between the Traps and Production Best Practices sections and the absence of any bundle files to offload the per-language example bulk.
Suggestions
Deduplicate guidance repeated between "Top Five Traps" and "Production Best Practices" (disposal, startup-vs-run failure distinction) — cover each once and cross-reference it from the checklist.
Consolidate the twice-stated note that inline MCP servers are not persisted across resume (Trap section and MCP Servers section) into a single location.
Trim the "Voice and Posture" section's avoid/prefer opener lists to a couple of illustrative examples, or move the full list to a references file.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body explains no background concepts Claude already knows, but carries real duplication: disposal is covered in Trap 3 and restated as Production Best Practice 1, the startup-vs-run-failure distinction is fully coded in Trap 2 and repeated as Best Practice 2, and the "inline MCP servers are not persisted across resume" note appears twice (lines 182 and 325). The ~20-line "Voice and Posture" opener-coaching list could also be tightened, making this more than the "minor instances" of anchor 4 but mostly efficient overall. | 3 / 5 |
Actionability | Every invocation pattern ships complete, copy-paste-ready code in both TypeScript and Python (one-shot `Agent.prompt`, `Agent.create` + `agent.send` with streaming loops, `Agent.resume`, full error handling with exit codes, `Cursor.models.list()`, `Agent.get`/`getRun` inspection), with specific option names like `skipReviewerRequest: true` and `autoCreatePR: true` covering the common cases. No pseudocode or gaps. | 5 / 5 |
Workflow Clarity | "Pick the language" is an explicitly ordered 1-2-3 decision procedure with an ask-and-wait checkpoint, error handling includes feedback loops (honor `isRetryable` / `retry_after` before exponential backoff), and "Production Best Practices" is an explicit 8-item checklist — matching the top anchor's sequence, recovery loops, and checklists. `supports(...)` guards and exit-code conventions close the checkpoint gaps that would justify anchor 4. | 5 / 5 |
Progressive Disclosure | Section structure is clean and the skill correctly defers full reference to the canonical docs, but no bundle files exist and all per-language code examples are inlined in a ~380-line body; some example bulk could live in `references/` files with clearly signaled links. Good structure with minor organization gaps fits anchor 4 rather than anchor 5, which requires well-signaled one-level-deep references. | 4 / 5 |
Total | 17 / 20 Passed |