Superpowers beats Agent Skills in head-to-head test: same app in 49 min for $4.76 vs 90 min for $6.30
chessbuzz · x · 2026-10-17
Gatherthink benchmarked two popular coding-agent frameworks by building the same 311 municipal application: Superpowers finished in 49 minutes for $4.76, while Agent Skills took about 90 minutes and cost $6.30. They then used both frameworks to add further features to the app.
Core argument: teams should pick an AI coding workflow now instead of waiting for Claude Code and Codex to build one in. Key points:
- A workflow makes agent coding repeatable and reviewable: same stages, same files left behind, same human sign-off points.
- A skill is one SKILL.md instruction file; a framework is a pack of skills forming a workflow. Claude Code (.claude/skills) and Codex (.agents/skills) read the same open skills format — commit the folder and every developer's agent shares the same playbook. Skills load on demand: agents only see name and description until needed.
- Superpowers installs a hook that reloads its rules on every session start, clear, or compaction.
- claude.md + scaffolding + plan mode can work alone, but frameworks add the repeatability and standards teams need; use all three, each covers a gap.
More from coding & agent
- Open-source Claude Code skill embeds a live website inside a tweet via player cards — doooyle · 2026-10-17
- YC Partner: Personal Software Is Cheaper Than Ever, and Open-Sourced Code Lets Others' Agents Fill the Gaps — ycombinator · 2026-10-17
- Claude Code's contextual UX hints earn praise from a heavy daily user — iamsahaj_xyz · 2026-10-17
- Codex-created worktrees blamed for killing a 600GB-battered SSD; users say disk bloat is the worst part of AI coding tools — natesiggard · 2026-10-17
- Engineer: I'm not smarter than AI anymore — time to learn to orchestrate it — KevinNaughtonJr · 2026-10-17
- Open-source OODA pentest agent drives Kali and Playwright tools directly over MCP — irraz_rulez · 2026-10-17