Release v1.4.0

What changed

• Added runtime generation controls:
  • --temperature
  • --max-tokens
  • --top-p
• Added optional runtime controls in TOML config (temperature, max_tokens, top_p) while keeping
  deterministic precedence.
• Extended normalized JSON metadata with execution observability:
  • metadata.execution_ms
  • optional metadata.usage (prompt_tokens, completion_tokens, total_tokens)
• Extended provider contract/runtime behavior to support:
  • provider-level control mapping from unified generation settings
  • optional normalized usage extraction across providers
• Updated technical documentation (README + docs) to reflect runtime controls, metadata behavior, and
  provider contract updates.

Why it matters

• Improves execution observability and operational control without changing the stateless
  architecture.
• Keeps one-shot deterministic behavior while exposing metrics useful for automation and cost/
  performance tracking.
• Preserves protocol-first provider design with normalized metadata for downstream tooling.

Stability Statement

• v1.4.0 is a stable minor release on top of the v1 baseline.
• No CLI breaking changes.
• No output contract breaking changes (metadata extension is additive).

Scope Clarification

• The project remains a deterministic stateless execution layer.
• No orchestration, no memory, no agents, no multi-turn workflow behavior.

Notes

• Full test suite passed before release packaging.
• Build artifacts and Twine checks passed for ai_prompt_runner-1.4.0.

GitHub Pull Requests Included

• feat(v1.4): add runtime controls contract and execution metadata by @damienSavoldelli in #18
• feat(v1.4): add provider runtime-controls mapping and usage normalization by @damienSavoldelli in #19
• docs(v1.4): document runtime controls and execution metadata by @damienSavoldelli in #20

Full Changelog: v1.3.0...v1.4.0