ouroboros/devtools
Andrei Kaznacheev 75c78ca20d Phase A: DeepSeek direct provider core (registry, transport, effort carriage, reasoning replay, tests)
- provider_models: deepseek:: prefix, DEEPSEEK_API_KEY, DEEPSEEK_BASE_URL constant,
  direct defaults (main/deep_self_review=v4-pro, light/fallback=v4-flash),
  main-x3 review roles, migrate/normalize branches, narrow vision prefix
- config/server_runtime/safety: exclusive-direct detection, startup gate,
  review fallback tuple, remote-provider keys, per-provider key env
- llm: deepseek target (carries_reasoning_effort), effort flag migration for
  direct OpenAI, ultra->max clamp, reasoning_content same-route replay +
  empty-fill (DeepSeek requires its own echo on tool turns), vision judged on
  qualified identity (fixes latent direct-lane blindness), cached_tokens
  top-level fallback
- secrets: masking/forbidden/colab/benchmark allowlists
- llm_probe dispatch set; pricing provider set
- tests: test_deepseek_provider.py (19 regressions) + registry pin update

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
2026-09-01 01:31:14 +04:00
..
benchmarks Phase A: DeepSeek direct provider core (registry, transport, effort carriage, reasoning replay, tests) 2026-09-01 01:31:14 +04:00
__init__.py feat(devtools-benchmarks): add official benchmark harnesses and workspace executor 2026-06-06 12:03:30 +03:00
README.md feat(devtools-benchmarks): add official benchmark harnesses and workspace executor 2026-06-06 12:03:30 +03:00

Ouroboros Devtools

devtools/ contains operator-side and benchmark support code that should be versioned with Ouroboros without becoming part of the runtime core.

Rules:

  • Generated logs, datasets, run outputs, Docker layers, and secrets do not live here.
  • Default benchmark outputs go under /Users/anton/Ouroboros/bench_runs/.
  • Runtime modules must not import devtools.
  • This is not an immune-system bypass: touched files are reviewed normally.
  • Promote code out of devtools only through a separate reviewed runtime plan.