Agent skill · lobehub
alint-ruleset
Maintain LobeHub's model-backed alint rule set in packages/alint: author a rule, calibrate its false positives against real source, decide its severity (warn vs error) from real PR data, and track its token cost. Use for adding or narrowing an alint rule, classifying findings as true/false positives, promoting or demoting a rule, or running the weekly rule-set review.
What it needs
About 4k tokens when loaded.
What this skill does
alint rule set packages/alint holds the model-backed lint rules for the judgement calls ESLint cannot express — semantic rules such as "fan-out over a runtime-sized list needs pMap" or "a secondary button is not primary". Each rule is a prompt that runs per file through alint and is cached by file content. This skill owns the lifecycle of that rule set: how a rule is written, how its false positives are removed, when it earns error, and what it costs. It does not own the rules themselves (they live in packages/alint/rules) and it does not restate the reading contract — that is packages/alint/README.md. Read that file once before editing a rule. The loop Everything below is one cycle. An ad-hoc task runs the phase it needs; the weekly review runs the incremental phases on a schedule. Phase What it does Reference ------------------------------ -------------------------------------------------------------------------------------------------- -------------------------------------------------------------- A · Collect the standard Pull the owner's recent review rejections, read them, keep the part that recurs and one file decides calibration.md B · Author the rule rule.alint.toml: what to report, where to anchor, and an explicit "do not report" list rule-authoring.md C · Calibrate Scan → sample → classify TP/FP → write the recurring FP patterns back as carve-outs → cold-run fixtures calibration.md D · Map the problems Cluster findings across rules by file/area; that is the cleanup backlog calibration.md E · Promote Read the rule's findings on real PRs; warn → error only at zero false positives promotion.md F · Account for the cost Token usage per run; cache invalidation is the cost driver cost.md Decision rules These are the calls that are easy to get wrong. They are not suggestions. The owner's words are the standard. A rule quotes the rejection it comes from, verbatim and in the original language; never paraphrase. …
How to use it
Reference it in AdaL, Claude Code, Cursor or any coding agent — nothing to install:
@skills lobehub/alint-ruleset--65e52e