cwc-long-running-agents vs loopy
Anthropic's harness primitives for long-running Claude agents: default-FAIL evidence gates, a fresh-context evaluator subagent and handoff hooks — each one standalone, readable file. — versus — A public library of reusable AI-agent loops plus Loopy, an installable skill that helps agents find, audit, adapt, run and publish loops from the live catalog.
Loopy catalogs reusable agent loops as installable skills; this repo ships the raw hook/evaluator primitives you'd build such loops from.
| cwc-long-running-agents | loopy | |
|---|---|---|
| Stars | 579 | 2.7k |
| Forks | 60 | 235 |
| Language | Shell | JavaScript |
| License | Apache-2.0 | MIT |
| Last activity | 2 months ago | 12 days ago |
| Topics | coding | skills |
| Curated connections | 3 | 3 |
cwc-long-running-agents — the curator's take
Read it, don't run it: an official worked example of WHY long runs fail (agents grading their own work, claiming success without evidence, losing state between sessions) with one hook per failure mode. The default-FAIL contract is the idea worth stealing. Explicitly an event demo — unmaintained, not accepting contributions — so copy the patterns into your own harness rather than depending on the repo.
loopy — the curator's take
The insight: most agent work is a repeatable loop someone already designed — so catalog them. The website is browsable by humans AND agents (llms.txt, JSON catalog, agent guide), and the Loopy skill turns your agent into a loop librarian: discover, audit, repair, debrief, publish. Genuinely useful for not reinventing the same research/review/refactor loop weekly. NOT a runtime — loops are prompts and procedure, not executable infrastructure; quality varies by contributor, so audit before you adopt (the skill's audit step exists for a reason).