ace-llm 0.30.0

data/CHANGELOG.md

@@ -0,0 +1,641 @@
+# Changelog
+
+All notable changes to this project will be documented in this file.
+
+The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/),
+and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
+
+## [Unreleased]
+
+## [0.30.0] - 2026-03-23
+
+### Added
+- Documented thinking levels (`low`, `medium`, `high`, `xhigh`) for CLI providers (`claude`, `codex`) in README and usage guide.
+- Documented built-in execution presets (`@ro`, `@rw`, `@yolo`) with supported CLI providers in README and usage guide.
+- Added `opus` and `haiku` aliases to the usage guide alias table.
+
+### Fixed
+- Updated stale Anthropic model aliases in `.ace-defaults/llm/providers/anthropic.yml` from Claude 3 IDs to Claude 4.5 (`claude-sonnet-4-5`, `claude-opus-4-5`, `claude-haiku-4-5`).
+
+### Changed
+- Rewrote README intro and Use Cases to highlight aliases, thinking levels, presets, and file-based prompts without repetition.
+- Replaced demo tape with focused queries showing `provider:model`, alias, thinking level, and preset calling patterns.
+- Updated gemspec description to include Gemini CLI and pi-agent.
+- Removed obsolete `docs/migration-from-llm-query.md` (internal-only migration guide for legacy `llm-query` tool).
+
+## [0.29.5] - 2026-03-23
+
+### Changed
+- Refreshed package README layout, quick-link navigation, and section flow to align with the current package pattern.
+
+## [0.29.4] - 2026-03-22
+
+### Changed
+- Replaced placeholder commands in `docs/demo/ace-llm-getting-started.tape.yml` with real `ace-llm` provider-listing and query flow commands.
+
+## [0.29.3] - 2026-03-22
+
+### Technical
+- Clarified release notes to document why the Codex `@yolo` preset intentionally omits `--full-auto` for current CLI compatibility.
+
+## [0.29.2] - 2026-03-22
+
+### Fixed
+- Clarified that the Codex `@yolo` preset omits `--full-auto` because the current Codex CLI rejects it alongside `--dangerously-bypass-approvals-and-sandbox`.
+
+## [0.29.1] - 2026-03-22
+
+### Fixed
+- Clarified the Codex `@yolo` preset docs after changelog drift described `--full-auto` as restored when it is intentionally omitted for CLI compatibility.
+
+## [0.29.0] - 2026-03-22
+
+### Changed
+- Rewrote the package README into a concise landing page and added focused documentation pages for getting started, usage, and handbook guidance.
+- Added a getting-started demo tape for the documented onboarding flow.
+- Updated gemspec summary and description to match the refreshed documentation surface.
+
+## [0.28.0] - 2026-03-21
+
+### Changed
+- Expanded `TS-LLM-001` E2E coverage with a deterministic unknown-provider routing goal and tightened verifier artifact evidence requirements for query and model-selection flows.
+
+## [0.27.1] - 2026-03-18
+
+### Changed
+- Migrated CLI namespace from `Ace::Core::CLI::*` to `Ace::Support::Cli::*` (ace-support-cli is now the canonical home for CLI infrastructure).
+
+
+## [0.27.0] - 2026-03-18
+
+### Changed
+- Removed legacy backward-compatibility behavior as part of the 0.10 cleanup release.
+
+
+## [0.26.13] - 2026-03-18
+
+### Added
+- Added Claude `@prompt` preset for strict prompt-only execution using verified no-tools mode.
+
+### Changed
+- Updated Claude `@ro` preset to allow `Bash` and `Read` tool access for richer read-only review workflows.
+- Clarified in usage docs that `@prompt` is currently provider-specific and only available where no-tools execution can be enforced.
+
+### Technical
+- Added regression coverage for Claude `@ro` and `@prompt` preset wiring.
+
+## [0.26.12] - 2026-03-18
+
+### Fixed
+- Changed Claude `@ro` preset behavior to disable tools without entering native Claude `plan` mode, preserving read-only execution while avoiding empty-response failures in one-shot JSON output.
+- Suppressed misleading `Total timeout exceeded (...)` status reporting after terminal single-provider failures so the real provider error remains the primary signal.
+
+### Technical
+- Added regression coverage for Claude `@ro` preset wiring and fallback terminal-error timeout reporting.
+
+## [0.26.11] - 2026-03-17
+
+### Fixed
+- Updated default provider context limits: Anthropic from 200K to 1M tokens, OpenAI from 128K to 1.05M tokens to reflect modern model capabilities.
+
+## [0.26.10] - 2026-03-17
+
+### Fixed
+- Resolve global alias inputs (for example `glite`) before provider/model validation so single-token alias-only provider inputs are accepted.
+
+### Technical
+- Fix `StubAliasResolver` test construction to pass hash as positional argument instead of keyword arguments.
+
+## [0.26.8] - 2026-03-17
+
+### Fixed
+- Parse explicit `:high`/`:low`/`:medium`/`:xhigh` thinking suffixes before alias resolution so model aliases with thinking modifiers (for example `codex:gpt:high@ro`) resolve correctly.
+
+## [0.26.7] - 2026-03-17
+
+### Fixed
+- Removed `--sandbox` from Gemini `ro` and `rw` presets; sandbox requires Docker/Podman and is orthogonal to `--approval-mode`.
+- Changed Gemini `rw` preset `--approval-mode` from invalid `auto` to `auto_edit` (valid choices: `default`, `auto_edit`, `yolo`, `plan`).
+
+## [0.26.6] - 2026-03-15
+
+### Changed
+- Migrated CLI framework from dry-cli to ace-support-cli
+
+## [0.26.5] - 2026-03-12
+
+### Added
+- Added `working_dir:` support to `QueryInterface.query` so CLI-backed provider executions can be anchored to an explicit filesystem root.
+
+## [0.26.4] - 2026-03-12
+
+### Changed
+- Updated the handbook LLM reference guide to describe canonical package `handbook/skills/` ownership before provider projections.
+
+### Fixed
+- Restore provider preset defaults under `.ace-defaults/llm/presets/` after the preset refactor dropped `max_tokens: 16000` and rewrote Codex/Gemini sandbox CLI flags; keep the newer 600-second timeout standardization intact.
+
+## [0.26.3] - 2026-03-07
+
+### Changed
+- Improve `--list-providers` output readability: show model count in provider header, wrap long model lists at 78 characters.
+
+## [0.26.2] - 2026-03-07
+
+### Changed
+- Refactor `Configuration#provider` to use `Enumerable#find` for idiomatic Ruby.
+- Use `<<~` HEREDOC for inactive-provider error message in `ProviderModelParser`.
+
+## [0.26.1] - 2026-03-07
+
+### Technical
+- Move `configuration_test.rb` to `test/organisms/` per ADR-017 flat test structure.
+
+## [0.26.0] - 2026-03-07
+
+### Added
+- Added provider allow-list filtering controls through `llm.providers.active` and `ACE_LLM_PROVIDERS_ACTIVE`, including normalization and env override handling.
+- Added configuration and parser tests covering allow-list behavior, env precedence, and inactive-vs-unknown provider classification.
+
+### Changed
+- Updated `ace-llm --list-providers` to show filtered-mode summary and inactive provider section when an allow-list is active.
+
+### Fixed
+- Distinguish inactive providers (configured but filtered out) from unknown providers in query validation errors, with actionable enablement guidance.
+
+## [0.25.1] - 2026-03-06
+
+### Added
+- Support explicit `:low`, `:medium`, `:high`, and `:xhigh` thinking suffixes on fully qualified model targets while preserving existing `@ro`, `@rw`, and `@yolo` preset behavior.
+- Add provider-scoped thinking overlays for Codex and Claude so explicit thinking levels map cleanly to provider-native runtime options.
+
+### Changed
+- Allow `ace-llm query` preset resolution to merge explicit thinking overrides after preset loading and before direct CLI overrides.
+
+## [0.25.0] - 2026-03-05
+
+### Added
+- `last_message_file:` parameter in `QueryInterface.query` to thread last-message file path through to provider clients via `generation_opts`.
+
+## [0.24.8] - 2026-03-04
+
+### Fixed
+- Normalize timeout inputs across `--timeout` flows in CLI query and fallback paths so string values are safely coerced to numeric before use.
+
+## [0.24.7] - 2026-03-04
+
+### Added
+- Plan-mode template at `tmpl://agent/plan-mode` (`ace-llm/handbook/templates/agent/plan-mode.template.md`) for reusable planning-only instruction composition.
+
+### Changed
+- Strengthen plan-mode template contract with explicit required section headings and stricter prohibitions against permission/escalation and status-only outputs.
+
+## [0.24.6] - 2026-02-28
+
+### Fixed
+- Resolve `provider:provider` format (e.g., `codex:codex`) to provider's default model instead of passing literal provider name as model; adds fallback in `ClientRegistry#resolve_alias` when model alias matches provider name
+
+## [0.24.5] - 2026-02-28
+
+### Fixed
+- Thread `--timeout` through fallback path: `FallbackOrchestrator` now accepts and forwards `timeout` to each `registry.get_client()` call, so `--timeout 300` is no longer silently dropped when fallback is enabled
+- Raise `max_tokens` defaults from 4096–8192 to 16384 across all gem providers (anthropic, openai, google, zai, groq, xai, mistral, togetherai, openrouter, lmstudio) and all project-level provider overrides in `.ace/llm/providers/`
+
+## [0.24.4] - 2026-02-27
+
+### Added
+- Per-provider fallback chains: `chains` map in config allows each primary provider to have its own contextual fallback order
+- `FallbackConfig#providers_for(primary)` returns chain-specific fallback list or default `providers`
+
+## [0.24.3] - 2026-02-27
+
+### Technical
+- Document `ZAI_API_KEY` in README API key section
+
+## [0.24.2] - 2026-02-27
+
+### Fixed
+- Add project-level fallback providers so `ace-git-commit` works without `ZAI_API_KEY`
+- Re-add CLI-specific error diagnostics listing available providers when provider not found
+- Tighten "window limit" quota-detection pattern to avoid matching "context window limit"
+
+### Technical
+- Add inline comment documenting per-layer normalization rationale in fallback config loading
+
+## [0.24.1] - 2026-02-27
+
+### Fixed
+- Narrowed quota-detection patterns to avoid false positives on unrelated error messages containing "credit" (e.g., "credentials invalid")
+
+## [0.24.0] - 2026-02-27
+
+### Added
+- Native `zai` API provider with direct HTTP calls and bearer auth (`ZAI_API_KEY`), supporting models `glm-4.7-flashx`, `glm-4.7`, and `glm-5`
+- Centralized `llm.fallback` config in `.ace-defaults/llm/config.yml` for shared fallback policy across QueryInterface and CLI callers
+
+### Fixed
+- Quota/credit/window-limit exhaustion classified as immediate fallback (no retry), while overload/unavailable/rate-limit conditions remain retryable
+- Fallback provider chains normalize and deduplicate entries (including aliases) while preserving order
+- Z.AI provider error surfacing includes actionable HTTP status and non-JSON response snippets
+
+### Changed
+- `ace-llm` CLI query command routes through `Ace::LLM::QueryInterface` for consistent fallback behavior across CLI and Ruby API
+- Fallback documentation updated to use `.ace/llm/config.yml` (`llm.fallback`) as primary configuration contract
+
+## [0.23.1] - 2026-02-23
+
+### Technical
+- Updated internal dependency version constraints to current releases
+
+## [0.23.0] - 2026-02-22
+
+### Changed
+- **Breaking:** Migrated from multi-command Registry to single-command pattern (task 278)
+  - Removed `query` subcommand: `ace-llm query gflash "prompt"` → `ace-llm gflash "prompt"`
+  - Removed `list-providers` subcommand: `ace-llm list-providers` → `ace-llm --list-providers`
+  - Removed `version`/`help` subcommands: use `--version`/`--help` flags only
+  - Added `--version` and `--list-providers` flags to main command
+  - Updated help text references from `ace-llm-query` to `ace-llm`
+  - No backward compatibility (per ADR-024)
+
+## [0.22.6] - 2026-02-22
+
+### Changed
+- Migrated from DefaultRouting to standard help pattern (task 278.21)
+  - Removed DWIM default routing to query command
+  - Added explicit `query` command requirement
+  - Removed `--list-providers` alias (use `list-providers` command)
+  - Added HelpCommand registration for `--help`, `-h`, `help`
+  - No-args now shows help instead of error
+
+### Technical
+- Updated CLI tests to use explicit `query` command prefix
+- Removed tests for deprecated DefaultRouting behavior
+
+## [0.22.5] - 2026-02-22
+
+### Fixed
+- Pass `backends` configuration from provider YAML to client, fixing ClaudeOaiClient missing env vars (`ANTHROPIC_BASE_URL`, `ANTHROPIC_AUTH_TOKEN`) and model tier mappings
+
+## [0.22.3] - 2026-02-21
+
+### Added
+- `subprocess_env:` parameter on `QueryInterface.query()` for passing environment variables to CLI provider subprocesses
+
+## [0.22.2] - 2026-02-17
+
+### Fixed
+- Added integration coverage to verify `QueryInterface.query(..., sandbox: ...)` forwards sandbox mode into generation options
+
+## [0.22.1] - 2026-02-15
+
+### Added
+- `sandbox:` parameter on `QueryInterface.query()` for controlling CLI provider sandbox mode (forwarded via `generation_opts`)
+
+## [0.22.0] - 2026-02-05
+
+### Added
+- `--cli-args` passthrough for CLI providers, with documentation and integration tests
+
+### Fixed
+- `--timeout` parsing in CLI query command to avoid string timeout errors
+
+## [0.21.1] - 2026-02-04
+
+### Added
+- Provider config now includes `context_limit` field for model context window sizes
+  - Google: 1M tokens (Gemini models)
+  - Anthropic: 200K tokens (Claude models)
+  - OpenAI: 128K tokens (GPT models)
+  - Other providers: 128K default
+- Default `context_limit` in main config.yml for unknown models
+
+### Technical
+- Lower Ruby version requirement to >= 3.2.0 across all gemspecs
+
+## [0.21.0] - 2026-01-14
+
+### Added
+- Configuration cascade for provider discovery
+  - Provider configs now cascade from gem defaults, project, and user paths
+  - Dynamic provider discovery without hardcoding in ace-llm
+  - New Configuration class and ConfigLoader molecule
+
+### Changed
+- Removed CLI provider configs from ace-llm gem (.ace-defaults/)
+- Updated ClientRegistry to use Configuration cascade
+
+### Fixed
+- Handle non-JSON xAI API error responses gracefully
+- Improved CLI argument parsing for ambiguous provider/model arguments
+
+## [0.20.2] - 2026-01-13
+
+### Fixed
+- Query command ambiguous argument handling when `--model` doesn't contain colon
+  - Added validation to detect when positional arg is not a valid provider
+  - Shows help instead of proceeding with invalid provider/model combination
+  - Fixed test to use non-alias value ("unknown-model" instead of "grok")
+
+## [0.20.1] - 2026-01-13
+
+### Fixed
+- Handle non-JSON xAI API error responses gracefully (task 205)
+  - Guard against String response bodies (e.g., HTML error pages from 502 gateway errors)
+  - Include HTTP status code in error messages for better debugging
+  - Apply safe pattern from OpenRouter client with explicit StandardError rescue
+
+## [0.20.0] - 2026-01-11
+
+### Changed
+- **BREAKING**: Renamed executable from `ace-llm-query` to `ace-llm` (task 202.01)
+  - The gem name remains `ace-llm`
+  - The executable is now `ace-llm` (previously `ace-llm-query`)
+  - Update scripts and documentation to use `ace-llm` command
+  - All functionality remains the same
+
+## [0.19.1] - 2026-01-09
+
+### Changed
+- **BREAKING**: Eliminate wrapper pattern in dry-cli commands
+  - Merged business logic directly into `ListProviders` and `Query` dry-cli command classes
+  - Deleted `list_providers_command.rb` and `query_command.rb` wrapper files
+  - Simplified architecture by removing unnecessary delegation layer
+
+## [0.19.0] - 2026-01-07
+
+### Changed
+- **BREAKING**: Migrated CLI framework from Thor to dry-cli (task 179.13)
+  - Replaced `thor` dependency with `dry-cli ~> 1.0`
+  - Created dry-cli command classes (query, list_providers)
+  - Standardized option type handling (temperature as float)
+
+### Fixed
+- CLI test compatibility with dry-cli capture patterns
+- Default command routing for query command
+
+## [0.18.0] - 2026-01-05
+
+### Added
+- Thor CLI migration with standardized command structure
+- ConfigSummary display for effective configuration with sensitive key filtering
+- Comprehensive CLI help documentation across all commands
+- Routing for list-providers command
+
+### Changed
+- Adopted Ace::Core::CLI::Base for standardized options (--quiet, --verbose, --debug)
+- Migrated from OptionParser to Thor framework
+- Added method_missing for default subcommand support
+
+### Fixed
+- CLI routing and dependency management for feature parity
+- --help dispatch for all ACE commands
+- Addressed PR #123 review findings for Medium and higher priority issues
+
+## [0.17.0] - 2026-01-03
+
+### Changed
+- **BREAKING**: Minimum Ruby version raised to 3.3.0 (was 3.1.0)
+- Standardized gemspec file patterns with deterministic Dir.glob
+- Added MIT LICENSE file
+
+## [0.16.1] - 2025-12-30
+
+### Changed
+
+* Add ace-config dependency for configuration cascade management
+* Refactor ClientRegistry to use Ace::Config.create() with deep merge
+
+## [0.16.0] - 2025-12-30
+
+### Changed
+
+* Rename `.ace.example/` to `.ace-defaults/` for gem defaults directory
+
+
+## [0.15.1] - 2025-12-14
+
+### Changed
+- **Standardized GENERATION_KEYS pattern**: All LLM clients now use declarative constants
+  - OpenAIClient, OpenRouterClient, GroqClient, MistralClient, AnthropicClient use `GENERATION_KEYS`
+  - GoogleClient uses `GENERATION_KEY_MAPPING` (maps internal keys to Gemini camelCase API keys)
+  - XAIClient already used this pattern (model implementation)
+  - Replaces inline nil checks with cleaner iteration loop
+  - Single point of truth for each client's supported parameters
+
+### Fixed
+- **MistralClient zero-value handling**: Fixed bug where `temperature: 0` was dropped
+  - Changed from truthiness checks to nil checks
+- **AnthropicClient zero-value handling**: Fixed bug where `temperature: 0` was dropped
+  - Changed from truthiness checks to nil checks
+- **GoogleClient zero-value handling**: Fixed bug where `temperature: 0` was dropped
+  - Changed from truthiness checks to nil checks
+
+## [0.15.0] - 2025-12-14
+
+### Added
+- **OpenAICompatibleParams module**: Shared concern for extracting OpenAI-compatible parameters
+  - Extracts `frequency_penalty` and `presence_penalty` using nil-safe logic
+  - Preserves zero values (uses `nil?` check, not truthiness)
+  - Included in XAIClient, OpenAIClient, and OpenRouterClient
+  - DRYs up code across OpenAI-compatible providers
+
+### Fixed
+- **OpenAIClient zero-value handling**: Fixed bug where zero-valued penalties were dropped
+  - `build_request_body` now uses nil checks for all generation parameters
+  - Zero values for `frequency_penalty`, `presence_penalty`, `temperature`, `max_tokens`, `top_p` are preserved
+- **YAML security hardening**: Removed Symbol from `permitted_classes` in `YAML.safe_load`
+  - Prevents potential DoS attacks via symbol table exhaustion
+  - All provider YAML configs already use string keys (no breaking changes)
+  - Added specific `Psych::DisallowedClass` warning for better debugging
+
+### Technical
+- Added comprehensive OpenAIClient test suite (15 new tests)
+- Added zero-value regression tests to XAIClient tests
+- Improved symbol rejection test with warning message assertions
+
+## [0.14.0] - 2025-12-07
+
+### Added
+- **Groq Provider**: New LLM provider for Groq's ultra-fast inference API
+  - Supports GPT-OSS 120B/20B, Kimi K2, and Mistral Saba models
+  - OpenAI-compatible API with ultra-fast inference
+  - Default model: openai/gpt-oss-120b with max_tokens: 4096
+  - Global aliases: `groq`, `groq-fast`, `groq-kimi`, `groq-saba`
+  - Model aliases: `gpt-oss`, `gpt-oss-120b`, `gpt-oss-20b`, `kimi-k2`, `saba`
+  - Environment variable: `GROQ_API_KEY`
+  - Supports `stop` sequences parameter
+
+### Fixed
+- **Zero-valued generation params**: Changed from truthiness checks to nil checks
+  - Allows `temperature: 0`, `frequency_penalty: 0`, `presence_penalty: 0`
+  - Affects GroqClient and ensures consistent behavior with OpenRouterClient
+
+### Changed
+- Stream flag explicitly disabled (streaming not implemented)
+- Parameter extraction refactored to use `Hash#slice` for cleaner code
+
+### Breaking Changes
+None
+
+## [0.13.0] - 2025-12-06
+
+### Added
+- **OpenRouter Provider**: New LLM provider for OpenRouter's unified API (400+ models)
+  - OpenAI-compatible API with optional attribution headers (HTTP-Referer, X-Title)
+  - Focus: Exclusive providers (DeepSeek, Kimi, Qwen) + fast inference via `:nitro` routing (Groq/Cerebras)
+  - Default model: openai/gpt-oss-120b:nitro with temperature: 0.7, max_tokens: 4096
+  - Fast inference aliases: `gpt-oss-nitro`, `kimi-nitro`, `qwen3-nitro`, `gpt-oss-small-nitro`
+  - Provider aliases: `deepseek`, `deepseek-r1`, `kimi`, `kimi-think`, `qwen-coder`, `qwq`, `hermes`, `glm`, `minimax`, `reka`, `devstral`
+  - Environment variable: `OPENROUTER_API_KEY`
+  - Handles non-JSON error responses (e.g., HTML from 502 errors)
+  - Explicit nil checks for generation params (allows temperature: 0, frequency_penalty: 0)
+  - Preserves `native_finish_reason` metadata from OpenRouter
+
+## [0.12.0] - 2025-12-06
+
+### Added
+- **x.ai (Grok) Provider**: New LLM provider for x.ai's Grok models
+  - Supports grok-4, grok-4-fast, grok-4-1-fast, grok-code-fast-1, grok-3, grok-3-fast, grok-3-mini, grok-2
+  - OpenAI-compatible API with full generation options
+  - Default model: grok-4 with max_tokens: 4096
+  - Global aliases: `grok` → xai:grok-4, `grokfast` → xai:grok-4-1-fast, `grokcode` → xai:grok-code-fast-1
+  - Environment variable: `XAI_API_KEY`
+
+### Changed
+- **Provider Config Migration**: Moved provider configs from `providers/` to `.ace.example/llm/providers/`
+  - Eliminates duplication between gem and project configs
+  - Example configs now serve as the canonical source
+- **XAIClient**: Uses constant-driven GENERATION_KEYS iteration pattern
+  - Explicit `rescue StandardError` instead of bare rescue
+
+## [0.11.0] - 2025-11-17
+
+### Added
+- **Graceful Provider Fallback**: Implemented automatic provider fallback with retry logic
+  - Automatic retry with exponential backoff (configurable, default 3 attempts)
+  - Intelligent error classification (retryable, skip to next, terminal)
+  - Fallback provider chain with configurable alternatives
+  - Total timeout protection (default 30s) to prevent infinite retry loops
+  - Jitter (10-30%) added to retry delays to prevent thundering herd issues
+  - Configurable via environment variables (`ACE_LLM_FALLBACK_*`) and runtime parameters
+  - Status callbacks for user visibility during fallback operations
+  - Respects Retry-After headers for rate limit compliance
+
+### Changed
+- **Fallback Orchestrator Refactoring**: Improved code organization and maintainability
+  - Extracted error handling logic into dedicated `handle_error` method for better separation of concerns
+  - Refactored `FallbackConfig.from_hash` with helper method to support both symbol and string keys
+  - Enhanced retry delay calculation with jitter to prevent synchronized retry storms
+  - Improved test coverage with range-based assertions for jittered delays
+
+### Technical
+- Added comprehensive test coverage for fallback system (atoms, molecules, models, integration)
+- Follows ATOM architecture pattern with clear separation: ErrorClassifier (Atom), FallbackConfig (Model), FallbackOrchestrator (Molecule)
+- Fixed minor style issues (missing newline at end of files)
+
+## [0.10.1] - 2025-11-16
+
+### Changed
+
+- **Dependency Update**: Updated ace-support-core dependency from `~> 0.9` to `~> 0.11`
+  - Provides access to latest PromptCacheManager features and infrastructure improvements
+  - Maintains compatibility with standardized ACE ecosystem patterns
+
+## [0.10.0] - 2025-11-15
+
+### Added
+- **System Prompt Control**: Added `--system-append` flag to ace-llm-query for flexible prompt composition
+  - `--system` flag now fully replaces provider defaults (clarified behavior)
+  - `--system-append` flag appends to existing/default system prompts
+  - Both flags can be used together for layered prompt control
+  - Claude provider: Maps to `--system-prompt` and `--append-system-prompt` CLI flags
+  - API providers (Anthropic, OpenAI, Google): Concatenates prompts with clear separator
+- **Enhanced Help Text**: Added provider-specific behavior notes to CLI help for system prompt flags
+  - Clarifies Claude uses native flags while API providers concatenate
+  - Improves user understanding of flag behavior across different providers
+
+### Fixed
+- **Claude Provider Bug**: Fixed ClaudeCodeClient to use correct `--system-prompt` flag instead of non-existent `--system`
+  - Resolves issue where system prompts were silently ignored with Claude
+  - Enables fast, deterministic responses with Claude Haiku for tools like ace-git-commit
+
+### Changed
+- **Code Organization**: Improved BaseClient helper method encapsulation
+  - Made helper methods (`concatenate_system_prompts`, `process_messages_with_system_append`, `deep_copy_messages`) private
+  - Relocated test file from `test/integration/` to `test/organisms/base_client_helpers_test.rb`
+  - Aligns with ACE flat test structure (tests are unit tests for organism helpers)
+- **Configuration**: Made system prompt separator configurable via `DEFAULT_SYSTEM_PROMPT_SEPARATOR` constant
+  - Addresses potential markdown conflicts in concatenated prompts
+  - Can be overridden by subclasses if needed
+- **Improved System Prompt Handling**: Refactored implementation with shared helpers
+  - Extracted concatenation logic to reduce code duplication
+  - Added safety checks and comprehensive test coverage (13 new tests)
+  - Implemented deep copy pattern to prevent message mutations
+
+### Technical
+- Added deprecation note for `append_system_prompt` option, prefer `system_append` for consistency
+
+## [0.9.5] - 2025-11-01
+
+### Changed
+
+- **Dependency Migration**: Updated to use renamed infrastructure gems
+  - Changed dependency from `ace-core` to `ace-support-core`
+  - Part of ecosystem-wide naming convention alignment for infrastructure gems
+
+## [0.9.4]
+ - 2025-10-08
+
+### Changed
+
+- **Test Structure Reorganization**: Reorganized tests for consistency
+  - Moved `test/ace/llm_test.rb` → `test/llm_test.rb`
+  - Moved `test/client_registry_test.rb` → `test/molecules/`
+  - Moved `test/provider_config_validator_test.rb` → `test/atoms/`
+  - Fixed require paths to match new structure
+  - Aligns with standardized flat ATOM structure across all ACE packages
+
+## [0.9.3] - 2025-10-07
+
+### Changed
+- **Test maintainability improvement**: Version tests now validate semantic versioning format instead of exact version values
+  - Prevents test failures on every version bump
+  - Uses regex pattern `/\A\d+\.\d+\.\d+/` to validate version format
+
+## [0.9.2] - 2025-10-07
+
+### Fixed
+- Standardized test file naming to follow project convention (`*_test.rb` suffix instead of `test_*` prefix)
+- Renamed `test_provider_config_validator.rb` → `provider_config_validator_test.rb`
+- Renamed `test_client_registry.rb` → `client_registry_test.rb`
+- Test count increased from 25 to 53 tests (all tests now properly discovered)
+
+## [0.9.1] - 2025-10-07
+
+### Added
+- `--model MODEL` flag for flexible LLM model specification in ace-llm-query CLI
+- Model resolution with priority: flag > positional > provider default
+- Dual syntax support: `PROVIDER[:MODEL]` and `PROVIDER --model MODEL`
+- `model:` parameter to `QueryInterface.query()` Ruby API for programmatic model override
+
+### Changed
+- CLI help banner updated to show both syntax options
+- README.md updated with `--model` flag documentation and examples
+
+## [0.9.0] - 2024-XX-XX
+
+### Added
+- Initial release with LLM provider integration
+- Unified CLI interface for querying multiple LLM providers
+- Support for Google Gemini, OpenAI, Anthropic, Mistral, Together AI, and LM Studio
+- Configuration-based provider architecture with YAML support
+- LLM alias resolution for quick access to models
+- Cost tracking and output formatting
+- ATOM architecture pattern (Atoms, Molecules, Organisms, Commands)
+
+
+## [0.22.4] - 2026-02-22
+
+### Fixed
+- Standardized quiet, verbose, debug option descriptions to canonical strings

data/LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2025 Michal Czyz
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.