evaluation

• 2026-04 agents.md effectiveness

  evaluating repository-level context files for coding agents.

• 2026-04 reward engineering for coding agents

  why coding agents optimize the rubric more than the prompt.

• 2026-04 reward hacking in coding agents

  how poorly designed metrics produce plausible but unstable code.

• 2026-04 preference toml

  use rlhf-shaped semantics in a simple config dsl for agent evaluation loops.

• 2026-04 reward rubric dsl

  a small machine-readable format for scoring coding agent output.