171,322 stars on a repo whose entire job is making your AI agent write less co… · @ReignSwords · TwKit

@ReignSwords

171,322 stars on a repo whose entire job is making your AI agent write less code. That's the whole pitch. Ponytail. One prompt. One KPI: shrink the diff. You ask an agent for a date picker. Without Ponytail, it hand-builds a calendar or installs a library: 335 lines. With it: 10 lines — because the repo already has an Input component and every browser ships a date picker. Their own benchmark (39 tasks, Opus 5.5, 5 runs each): 1. -53% code, -41% time, -26% cost, -45% tokens 2. 98% of risky logic shipped with a test — vs 68% without the skill 3. Security tasks (SQLi, path traversal, forged tokens, rate limiting): 30/30 passed The part that actually matters: less code didn't mean less safe. The rule was never "fewest tokens." Validation, error handling, and security are never on the chopping block. The code ends up small because it's necessary, not golfed. The philosophy is a ladder, applied before writing anything: 1. Does this need to exist? 2. Is it already in this codebase? 3. Does the standard library do it? 4. A native platform feature? 5. An installed dependency? 6. Can it be one line? 7. Only then: the minimum that works, plus one small test if it has logic. My take: every other agent skill on GitHub is teaching your agent new tricks. This is the only one whose job is deleting the tricks it doesn't need. Even `/ponytail-debt` harvests your `shortcut:` comments into a ledger so "later" doesn't become "never." Cold water, because I read the README: - The benchmark is vendor-run, 39 tasks. Real, but small. No independent replication yet. - No prompt survives an agent that really wants to write that 120-line cache class. The README admits it: "Insist anyway and he'll build it. While looking at you." - Adapter sprawl: 20 plugin dirs (Claude Code, Codex, Cursor, Copilot, OpenCode, Gemini...). The skill is thin; the adapters are the maintenance burden. Who should care: anyone paying per token, or reviewing agent PRs at 11pm. Who shouldn't: teams where writing boilerplate is someone's job security. Repo in the first comment. If your agent wrote 300 lines yesterday — how many would survive this ladder? Be honest. If you're the kind of dev who deletes code on Fridays, we're already friends. 🔵 Fellow Premium members, let's connect! I post about AI & tech — follow for more 👥

曝光 69 · 评论 6 · 点赞 0 · 书签 0