Foreman 0.5.1

Foreman 0.5.1

This release adds provider-independent decision models and fixes managed Codex worker supervision.

• Select Jev or OpenAI for responsibility routing and semantic checks through
  FOREMAN_DECISION_PROVIDER and FOREMAN_DECISION_MODEL. The OpenAI adapter uses the native
  Decisions API. Install foreman-core[openai] to include its SDK.
• Existing Jev configuration and public compatibility facades continue to work. Generic decision
  request budgets and provider/model usage metadata support both adapters.
• Keep decision credentials scoped to the supervisor, including credentials loaded from central
  configuration. Existing worker credentials and model settings retain their own scope.
• Set FOREMAN_CODEX_EXECUTABLE or FOREMAN_CODEX_MODEL to explicitly configure Codex coding
  and verifier workers. Defaults still use the installed CLI and native model configuration.
  Unsupported worker-model errors escalate with guidance; Foreman does not silently pick a fallback.
• Prevent native Foreman hooks from supervising a factory-owned worker a second time. Normalize
  completed App Server output so actual test results and agent conclusions remain readable, and
  include explicit independent-verification lifecycle evidence. Evidence compaction reduces bulky
  transcripts before cutting source or test evidence. Completion thresholds are unchanged.

See decision-model configuration and
worker configuration for setup and troubleshooting.

Install or upgrade:

python -m pip install --upgrade foreman-core==0.5.1
# Include the OpenAI decision adapter SDK:
python -m pip install --upgrade "foreman-core[openai]==0.5.1"