Content
82%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 high-quality, repo-anchored review skill: it gives concrete helpers, exact SQL/pattern shapes, red flags, fixes, and a report format, and the workflow is well sequenced with a final go/no-go gate. Remaining gains are modest: deduplicate Red Flags/Preferred Fixes against the workflow steps, add a post-finding feedback step, and move the repo-specific helper catalog and KB-connector detail into reference files.
Suggestions
Move the repo-specific helper catalog ("Sim Helpers To Prefer") and the KB Connector File Size Handling deep-dive into one-level-deep reference files (e.g., references/helpers.md, references/kb-connector.md), keeping a brief pointer plus the litmus test in SKILL.md.
Deduplicate Red Flags and Preferred Fixes against the Review Workflow steps (e.g., the OFFSET/pagination and Promise.all items appear in both) to tighten conciseness.
Add a feedback step after the Findings Format — e.g., re-verify each P1 finding against the changed code paths and confirm the good-to-go gate before reporting — to complete the workflow loop.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense and assumes competence — no concept-explaining filler, and everything is either a repo-specific fact or a check Claude could not know. Minor trims exist: Red Flags / Preferred Fixes partially restate Review Workflow checks, and the KB-connector section runs long with some repetition of the Sim Helpers entry. | 4 / 5 |
Actionability | Fully concrete guidance for an instruction-only skill: named helpers (chunkedBatchDelete, readResponseToBufferWithLimit, selectRowsByIdChunks), real file paths, an exact SQL shape ("WHERE id > afterId ORDER BY id LIMIT N"), and a copy-paste findings report template with a worked P1 example. | 5 / 5 |
Workflow Clarity | A clearly sequenced 7-step review workflow where each step carries its own verification question (e.g., step 2: "If the code does not enforce one, it is unbounded"), plus a final explicit gate ("Only say 'good to go' when every changed source has explicit row, byte, and concurrency bounds"). Not 5 because there is no feedback loop after findings are drafted (re-check, re-run, or escalation guidance). | 4 / 5 |
Progressive Disclosure | No bundle files exist and the single document is cleanly sectioned and navigable; however, the repo-specific helper catalog ("Sim Helpers To Prefer") and the ~20-line KB-connector deep-dive are strong candidates for one-level-deep reference files, so the main file carries detail that could be split out — good structure with minor organization gaps. | 4 / 5 |
Total | 17 / 20 Passed |