Popular repositories Loading

1. sidekick-harness sidekick-harness Public

   When does delegating to a cheap sidekick model actually pay? A minimal replication of Cognition's lead/sidekick architecture — cost/score pairs, delegation-behavior metrics, and full transcripts fo…

   Python

2. review-agent-lab-public review-agent-lab-public Public

   When does context help a code-review agent? A measured answer — seeded-bug evalset, strict audited scoring, every claim recomputed from committed run records in CI.

   Python

3. rl-environments rl-environments Public

   Agent-evaluation environments: planted-truth worlds, ungameable graders, calibrated difficulty

   Python

4. evalplane evalplane Public

   A distributed evaluation platform for coding agents. Refuses invalid comparisons and reports when a difference is inside the noise floor.

   Go