mpfaffenberger/code_puppy

Added

• Add live integration test for summarization failure → truncation fallback
• Add Wafer.AI model definitions to models.json with updated context length
• Add JSON‑string argument parsing support and coverage tests for universal_constructor

Changed

• Calibrate token estimates for opus‑4‑7 models and implement auto‑fallback to truncation on summarization errors
• Update universal_constructor imports to include Union and refactor implementation
• Bump project version to 0.0.468 and update lockfile

Fixed

• Skip SYN_API_KEY placeholder values in CI when gating live integration tests
• Raise MAX_HEADER_LENGTH limit to 60 characters in ask_user_question validation

Added

• Add non‑streaming render fallback, exception diagnostics, and retryable streaming classifier modules
• Introduce prepare_model_prompt hook and claude‑code prompt plugin
• Add summarization_model configuration option for custom summarization model

Changed

• Refactor BaseAgent into modular submodules (_builder, _compaction, _history, _key_listeners, _runtime)
• Bump termflow‑md dependency to >=0.1.11
• Update invoke_agent to persist partial session history on mid‑run crashes

Removed

• Remove antigravity OAuth plugin and MOTD feature with all related code and hooks

Fixed

• Fix agent_tools invoke_agent crash caused by missing load_puppy_rules method
• Fix diff color normalization, replace_in_file argument coercion, and fast‑mode config handling
• Fix compaction deferral bug for Claude Opus 4.7 by unifying tool part classification

Added

• Add support for Claude Opus 4.7 model family across CLI runner, core commands, OAuth plugin, and model utilities
• Introduce transient streaming retry logic with exponential backoff for agent run requests

Changed

• Bump project version to 0.0.459 in pyproject.toml and uv.lock

Fixed

• Correct should_use_anthropic_thinking_summary check to recognize "opus-4-7" model names

Added

• Introduce configurable max_hook_retries via puppy.cfg with get_max_hook_retries()
• Add Anthropic Open 4.7 thinking summary flag via should_use_anthropic_thinking_summary()

Changed

• Bump package version to 0.0.457 in pyproject.toml and uv.lock

Fixed

• Remove stray DBOS.step decorator from message_history_accumulator
• Pass usage_limits argument to _run_with_result_hooks to resolve undefined name error

Added

• Introduce agent_run_result lifecycle hook with on_agent_run_result callback and integrate into BaseAgent run flow

Changed

• Bump package version to 0.0.456