v0.5.0

What's Changed

Other Changes

• feat: add swebench-verified and terminal-bench tasks by @BobbyZhouZijian in #61
• refactor: load graders via entrypoint + isolated venv by @BobbyZhouZijian in #63
• docs/init: clean up stale grader fields after #63 by @BobbyZhouZijian in #64
• fix: agent-context coral cmds respect .coral_dir breadcrumb by @BobbyZhouZijian in #65
• refactor: migrate dna_design, swebench-verified, terminal-bench to entrypoint graders by @BobbyZhouZijian in #66

Full Changelog: v0.4.0...v0.5.0

v0.4.0

What's Changed

Other Changes

• feat: add file organizer and deep research skills by @BobbyZhouZijian in #52
• fix: use YAML arrays for tools/skills in agent frontmatter by @BobbyZhouZijian in #53
• fix: use record format for tools/skills in agent frontmatter by @BobbyZhouZijian in #56
• feat: auto-restart stalled agents via timeout by @BobbyZhouZijian in #57
• fix: remove unused research_turns from warmstart config by @BobbyZhouZijian in #58
• fix: only use sudo for docker when necessary by @BobbyZhouZijian in #60
• feat: async grader daemon for eval pipeline by @BobbyZhouZijian in #62

Full Changelog: v0.3.0...v0.4.0

v0.3.0

What's Changed

Other Changes

• Feat add frontier cs examples and fix agent permission bugs by @BobbyZhouZijian in #2
• fix the permission for codex and claude code by @BobbyZhouZijian in #4
• Add runtime-specific Codex options via agents.runtime_options by @YihongT in #3
• Add more math problems by @BobbyZhouZijian in #5
• Add better cli by allowing selecting which run to operate on by @BobbyZhouZijian in #8
• fix: resolve tmux breadcrumb path relative to CWD by @BobbyZhouZijian in #9
• fix: cap Python version at <3.14 for ML package compatibility by @BobbyZhouZijian in #10
• fix: pass grader.timeout to TaskGrader instances by @BobbyZhouZijian in #13
• feat: per-worktree dependency isolation and robust grader execution by @BobbyZhouZijian in #11
• fix: --full-auto sandbox blocks git worktree writes and CUDA by @BobbyZhouZijian in #14
• fix: pass task/run to inner resume command to avoid double prompt in tmux by @BobbyZhouZijian in #19
• feat: migrate config to OmegaConf with dotlist CLI overrides by @BobbyZhouZijian in #16
• fix: sort notes by timestamp instead of filename order by @BobbyZhouZijian in #17
• feat: add --instruction flag to coral resume by @BobbyZhouZijian in #23
• fix: persist config overrides to disk on resume by @BobbyZhouZijian in #24
• Fix venv setup race condition with file lock by @BobbyZhouZijian in #25
• Fix: isolate per-worktree venvs with UV_PROJECT_ENVIRONMENT by @BobbyZhouZijian in #26
• feat: add --diff flag to coral show by @BobbyZhouZijian in #27
• feat: improve heartbeat prompts by @BobbyZhouZijian in #20
• fix: use unique tmux session names so concurrent runs coexist by @BobbyZhouZijian in #29
• fix: normalize naive datetimes to UTC in notes sort key by @MikeZheng777 in #32
• feat: add Kiro CLI runtime + agent spawn stagger for rate-limit backpressure by @Xuan-1998 in #31
• feat: add OpenCode workspace guard and fix naive datetime sort by @MikeZheng777 in #35
• feat: add Docker session mode with refactored helpers by @BobbyZhouZijian in #36
• feat: add LiteLLM gateway for intercepting agent model traffic by @BobbyZhouZijian in #34
• add cc litellm config by @BobbyZhouZijian in #38
• Add paper link by @BobbyZhouZijian in #39
• update citation bibtex by @BobbyZhouZijian in #40
• Fix multiple mounts and codex setup with litellm by @BobbyZhouZijian in #41
• Fix venv resolution issues by @BobbyZhouZijian in #42
• Remove obsolete task.seed and task.files config fields by @BobbyZhouZijian in #44
• feat: add warm-start research phase by @MikeZheng777 in #45
• fix: base url field by @BobbyZhouZijian in #46
• feat: Add DNA enhancer and antibiotic drug design examples from SAGA by @BobbyZhouZijian in #48

New Contributors

• @YihongT made their first contribution in #3
• @MikeZheng777 made their first contribution in #32
• @Xuan-1998 made their first contribution in #31

Full Changelog: https://github.com/Human-Agent-Society/CORAL/commits/v0.3.0