Skip to content

refactor: move 15 benchmark DomainConfigs out of agent/domain.py to configs (CODE_HEALTH #3) #40

Description

@duanyiqun

CODE_HEALTH #3 (see docs/CODE_HEALTH.md in #37).

src/cognifold/agent/domain.py (1,500 lines) defines 21 module-level DomainConfig literals — 15 of them benchmark-only (LOCOMO, FUTUREX, LONGMEMEVAL, MSC, BABILONG, MUTUAL, MUSIQUE, STREAMINGQA, RGB, TIMEQA, NARRATIVEQA, QMSUM, SOCIALIQA, TOMI, SAFETYBENCH; lines 419–1310), several for benchmarks whose code was removed from the repo. The extension point register_domain() (line 1494) exists and is unused for these. The list has already drifted out of sync with query/agent.py's benchmark dispatch (TIMEQA has a domain but no query handler).

Fix direction:

  • Keep the ~6 real product domains in-tree
  • Move benchmark domains to YAML under configs/ (which already hosts the per-benchmark prompt profiles)
  • Each runner registers its domain via register_domain() at startup
  • Delete domains for removed benchmarks outright

Pairs naturally with #39 (query/agent.py de-benchmarking).

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions