Raimondas
@raimondas.bsky.social
Principal Data Scientist at Microsoft working on agentic systems and enterprise AI. Opinions and posts my own.
2,495 tests. Four rounds of hostile review. ~203 defects.
My test suite discovered none of them.
A suite is a ratchet, not a detector. It can't fail for a bug nobody has thought of, because every test is a written-down thought.
github.com/raimondasl/u...
My test suite discovered none of them.
A suite is a ratchet, not a detector. It can't fail for a bug nobody has thought of, because every test is a written-down thought.
github.com/raimondasl/u...
ultracua/docs/method.md at main · raimondasl/ultracua
Contribute to raimondasl/ultracua development by creating an account on GitHub.
github.com
September 11, 2026 at 3:18 PM
2,495 tests. Four rounds of hostile review. ~203 defects.
My test suite discovered none of them.
A suite is a ratchet, not a detector. It can't fail for a bug nobody has thought of, because every test is a written-down thought.
github.com/raimondasl/u...
My test suite discovered none of them.
A suite is a ratchet, not a detector. It can't fail for a bug nobody has thought of, because every test is a written-down thought.
github.com/raimondasl/u...
69% of my coding agent's "done, tests pass" claims had no passing test run behind them (591 sessions, checked from the transcripts). So: a Stop hook that runs the repo's real tests before the agent can say done. No LLM, no deps, MIT. Claude Code, Codex, Cursor, Gemini CLI. github.com/raimondasl/i...
GitHub - raimondasl/isitdone: Don't let your coding agent say "done" until the tests actually pass. Zero-LLM Stop hook + CLI for Claude Code, Codex, Cursor and Gemini CLI.
Don't let your coding agent say "done" until the tests actually pass. Zero-LLM Stop hook + CLI for Claude Code, Codex, Cursor and Gemini CLI. - raimondasl/isitdone
github.com
September 9, 2026 at 3:31 PM
69% of my coding agent's "done, tests pass" claims had no passing test run behind them (591 sessions, checked from the transcripts). So: a Stop hook that runs the repo's real tests before the agent can say done. No LLM, no deps, MIT. Claude Code, Codex, Cursor, Gemini CLI. github.com/raimondasl/i...