Ponytail 5 is rebuilt from the ground up: new rules, a new review, a new audit.
- Rules: rewritten and about half as long. The goal is the smallest complete change: before it writes, the agent maps everything the change must reach (callers, tests, config) and what it could break. Every reply ends with what it skipped or did not check, and any risk.
- Codebase map: at session start the agent gets a short map of what already exists in the repo, so it reuses before it writes.
/ponytail-review: rebuilt as a full review: bugs, security, real load, missing tests, speed, and what to cut. It reads the code the change touches, not just the diff. Each finding says what the code does, what goes wrong, how to fix it, and what happens if you skip it./ponytail-audit: the same checks for the whole repo, ranked, with what to fix first.
Against the same agent without the skill (Claude Code, Opus 5.5, 39 tasks including a real FastAPI + React repo, 5 runs each): -53% code, -41% time, -26% cost, -45% output tokens, and 98% of risky logic ships with a test (without: 68%). Method and limits: benchmarks/results/2026-10-07-agentic.md.
What's Changed
- chore: remove unused ballast found in the repo inventory by @DietrichGebert in #1042
- chore: move translations, CONTRIBUTING and .env.example out of the repo root by @DietrichGebert in #1044
- feat: Ponytail 5, rebuilt rules, review and audit by @DietrichGebert in #1061
- chore: release v5.0.0 by @DietrichGebert in #1062
Full Changelog: v4.13.0...v5.0.0