I rejected a 286k-star framework but kept one rule — and it's the only part still running a month later
Frequent-Ad-836 · reddit · 2026-09-14
The author evaluated obra/superpowers (286k stars), skipped installing it, but adopted one rule from its subagent-driven-development skill: a running plan doesn't wait on humans. Agents decide conflicts themselves, recording each as Ruling: decision — why — cost if wrong, with only four stop conditions (irreversible ops, security-sensitive actions, side effects outside the worktree, hopelessly broken plans). They added a "Decisions I made on your behalf" section to every unattended run's completion report — printing "none" when empty so emptiness and omission are distinguishable. After a month, decision count unchanged, but rulings moved from unread JSON fragments to one block read first after each run. Key insight: the value is collecting decisions before their context dies, not recording them. They skipped the framework itself for injecting a 63-line non-negotiable bootstrap that would clash with their own rule set.
More from coding & agent
- Agent got rescued by hand? Route that fix back into skill versioning — Jimcy-Maffesoli · 2026-09-14
- Property manager: Meta's Muse agent handled a 2am sewage backup end-to-end — armand_ruiz · 2026-09-14
- Memanto: open-source memory agent managing AI agents' memory, 2.2k stars — Shruti_0810 · 2026-09-14
- MAVIS: a personal agent that writes and tests its own tools and sub-agents — trinitron1f · 2026-09-14
- Anthropic demos Claude as on-call engineer: 15-minute incident triage in Slack — xiaohu · 2026-09-14
- Agent hacks stem from scraping local private files, not alignment failure, researcher argues — ryunuck · 2026-09-14