Back to Oh My Openagent

Task 1 evidence - builtin curated agent fallback chains

.omo/evidence/omo-senpi-adapter/20260719-curated-agents/task-1.md

4.19.1988 B
Original Source

Task 1 evidence - builtin curated agent fallback chains

What was tested

  • The five mirrored chains are keyed only by explore, librarian, metis, momus, and oracle.
  • The test contains an independent literal transcription of every provider list, model, variant, and array position. It does not import @oh-my-opencode/model-core.
  • Mutation proof: changing the non-head Explore Qwen entry while retaining the same chain length failed the literal table assertion; the mutation was reverted.

What was observed

  • bun test packages/senpi-task/src/agents/builtin/fallback-chains.test.ts completed with 5 pass and 0 fail.
  • The Oracle head remains the hand-transcribed GPT-5.5 high entry required by the plan.

Why it is enough

The complete independent value table fails on entry drift that length-only checks cannot detect, while keeping the package decoupled from model-core.

What was omitted

Runtime registry resolution belongs to task 4 and is covered there.