Tomorrow we launch Replume on Product Hunt!
Know what your coding agents actually did: every step of Claude Code, Codex and Cursor checked, tests rerun by Replume itself, evidence on the PR.
producthunt.com/products/rep…
Two releases this week: replume-cli 0.2.3 and 0.2.4.
What you see in the video: the agent fixes the code and says done; Replume reruns the tests itself and writes the report.
What's new under it:
→ blocks a test that was bent to fit the bug (red test, expected value changed, code untouched)
→ commit gate: no git commit or push after a change until the tests ran green
→ catches weakened jest/vitest/pytest config and removed asserts; INCOMPLETE if the suite shrank
→ every stop says why and what to do next
→ replume status, uninstall, report --agent
Claude Code verified in live sessions; Codex early.
npm install -g replume-cli
replume now works with GitHub Copilot and Hermes too, on top of Claude Code, Codex and Cursor!
which agent should we add next?
replume.dev
Whatever the prompt says, replume runs the tests when the agent says done.
To show it, we told Claude Code not to. It fixed a file, said done. Replume ran them: one failed, agent sent back. Rerun: 5/5.
What do you check by hand after "done"?
replume.dev
We asked Claude Code to rename a test. Replume's hook blocked it: that file is protected by the task contract, so the agent stopped and asked first. Agents shouldn't be able to make tests pass by editing the tests. Early build. Waitlist: replume.dev



