GiviLoop: an open-source workflow where Claude's code review finds one real bug and one false finding gets dismissed
Smart_Ad500 · reddit · 2026-09-26
The author of GiviLoop, a free MIT-licensed CLI/MCP tool, shares a real example from its own development: Claude's code review caught a genuine bug (an unknown verdict in saved findings could be omitted from report counts), reproduced by a test and fixed; a second claim about an unhandled JSON error was tested against actual CLI behavior and dismissed.
Workflow highlights:
- Have the coding agent verify each review finding against code and tests, persisting confirmed/dismissed/unverified findings with review IDs and evidence
- Context is auto-prepared from Git and files, avoiding long handoffs
- Works with local models or existing web chats, avoiding a separate paid review API (total savings unmeasured)
- Project and report example on GitHub (vgflutter/GiviLoop)
More from coding & agent
- One agent writes the fix, another reviews it: a two-agent code review workflow in Slack — Al_Grigor · 2026-09-26
- Academic agent Memex upgraded to Opus 5.5: writing quality fixed, experience much better — arjunrajlab · 2026-09-26
- Dev uses open-source Ling-3.0-flash-VL to let AI redesign the foldable iPhone in a single HTML file — alifcoder · 2026-09-26
- Anthropic launches Claude plugin directory portal as MCP usage jumps 110x this year — ClaudeDevs · 2026-09-26
- Open-source Jev agent plays Pokemon Red live, pushing fast-decision AI beyond Tetris — supportingthedogs · 2026-09-26
- Anthropic deep dive: effort tuning in Claude Code pays off most for security and code review — trq212 · 2026-09-26