ace-llm-providers-cli 0.27.0

checksums.yaml

@@ -0,0 +1,7 @@
+---
+SHA256:
+  metadata.gz: 5065e45a6f893a5a089d59ea40e1ccbbbe009bd0863b60ed0410f924073a8024
+  data.tar.gz: 90e8f0276496a9812407682c11f91da03f8abaa424d20ec6dad4864817a03590
+SHA512:
+  metadata.gz: 500f6b58d1bf63596f0e9c279f7fb0523ba942e30b3160f2ec9201f4aac4cebf6ac1a0db35a1677a3cae089fa0569b0c5f8796ccff4872247e15b4150e75ffac
+  data.tar.gz: 14d49ef076e20414f53d896c51c8b9e046a478da069e5978c85168bd24275c9970462f21547ef69faa10822560e950be5f62a1f21bdc64bd55c61540791c9897

data/.ace-defaults/llm/providers/claude.yml

@@ -0,0 +1,24 @@
+name: claude
+class: Ace::LLM::Providers::CLI::ClaudeCodeClient
+gem: ace-llm-providers-cli
+models:
+  - claude-opus-4-1
+  - claude-sonnet-4-0
+  - claude-3-5-haiku-latest
+aliases:
+  global:
+    cc: claude:sonnet
+    cc-opus: claude:opus
+    cc-sonnet: claude:sonnet
+    cc-haiku: claude:haiku
+  model:
+    opus: claude-opus-4-1
+    sonnet: claude-sonnet-4-0
+    haiku: claude-3-5-haiku-latest
+capabilities:
+  - text_generation
+  - streaming
+  - function_calling
+  - vision
+default_options:
+  temperature: 0.7

data/.ace-defaults/llm/providers/codex.yml

@@ -0,0 +1,22 @@
+name: codex
+class: Ace::LLM::Providers::CLI::CodexClient
+gem: ace-llm-providers-cli
+models:
+  - gpt-5
+  - gpt-5-mini
+aliases:
+  global:
+    codex-gpt5: codex:gpt-5
+    codex-mini: codex:gpt-5-mini
+  model:
+    5: gpt-5
+    mini: gpt-5-mini
+    m: gpt-5-mini
+capabilities:
+  - text_generation
+  - code_generation
+  - function_calling
+  - skill_rewriting
+default_options:
+  temperature: 0.7
+  skills_dir: .codex/skills

data/.ace-defaults/llm/providers/codexoss.yml

@@ -0,0 +1,13 @@
+name: codexoss
+class: Ace::LLM::Providers::CLI::CodexOSSClient
+gem: ace-llm-providers-cli
+models:
+  - default
+aliases:
+  global:
+    codex-oss: codexoss:default
+capabilities:
+  - text_generation
+  - code_generation
+default_options:
+  temperature: 0.7

data/.ace-defaults/llm/providers/gemini.yml

@@ -0,0 +1,32 @@
+name: gemini
+class: Ace::LLM::Providers::CLI::GeminiClient
+gem: ace-llm-providers-cli
+models_dev_id: google
+models:
+  - gemini-2.5-flash
+  - gemini-2.5-pro
+  - gemini-2.0-flash
+  - gemini-1.5-pro-latest
+aliases:
+  global:
+    gemini-flash: gemini:gemini-2.5-flash
+    gemini-pro: gemini:gemini-2.5-pro
+    gflash: gemini:gemini-2.5-flash
+    gpro: gemini:gemini-2.5-pro
+  model:
+    2.5-flash: gemini-2.5-flash
+    2.5-pro: gemini-2.5-pro
+    2.0-flash: gemini-2.0-flash
+    1.5-pro: gemini-1.5-pro-latest
+    flash: gemini-2.5-flash
+    pro: gemini-2.5-pro
+capabilities:
+  - text_generation
+  - code_generation
+  - vision
+default_options:
+  temperature: 0.7
+  # Maximum prompt length before switching to file-based prompts (characters)
+  # Gemini's actual token limit is ~1M, but this provides a safeguard for CLI usage
+  max_prompt_length: 100000
+  system_prompt: "Output only the direct answer to stdout. No conversational filler, no acknowledgments, no explanations. Do not attempt to use tools. Be concise and direct."

data/.ace-defaults/llm/providers/opencode.yml

@@ -0,0 +1,26 @@
+name: opencode
+class: Ace::LLM::Providers::CLI::OpenCodeClient
+gem: ace-llm-providers-cli
+models:
+  - google/gemini-2.5-flash
+  - google/gemini-2.0-flash-experimental
+  - google/gemini-1.5-pro
+  - anthropic/claude-3-5-sonnet
+  - anthropic/claude-3-5-haiku
+  - openai/gpt-4o
+  - openai/gpt-4o-mini
+aliases:
+  global:
+    oc: opencode:google/gemini-2.5-flash
+  model:
+    gemini: google/gemini-2.5-flash
+    claude: anthropic/claude-3-5-sonnet
+    gpt4: openai/gpt-4o
+capabilities:
+  - text_generation
+  - streaming
+  - function_calling
+  - vision
+  - multi_provider
+default_options:
+  temperature: 0.7

data/.ace-defaults/llm/providers/pi.yml

@@ -0,0 +1,43 @@
+name: pi
+class: Ace::LLM::Providers::CLI::PiClient
+gem: ace-llm-providers-cli
+models:
+  # ZAI (default provider)
+  - zai/glm-4.7
+  # Anthropic
+  - anthropic/claude-opus-4-6
+  - anthropic/claude-opus-4-5
+  - anthropic/claude-sonnet-4-5
+  - anthropic/claude-sonnet-4-0
+  - anthropic/claude-haiku-4-5
+  # Google Gemini
+  - google-gemini-cli/gemini-3-pro-preview
+  - google-gemini-cli/gemini-3-flash-preview
+  - google-gemini-cli/gemini-2.5-pro
+  - google-gemini-cli/gemini-2.5-flash
+  # OpenAI Codex
+  - openai-codex/gpt-5.3-codex
+  - openai-codex/gpt-5.2-codex
+  - openai-codex/gpt-5.2
+  - openai-codex/gpt-5.1
+aliases:
+  global:
+    pi: pi:zai/glm-4.7
+  model:
+    glm: zai/glm-4.7
+    opus: anthropic/claude-opus-4-6
+    sonnet: anthropic/claude-sonnet-4-5
+    haiku: anthropic/claude-haiku-4-5
+    gemini-pro: google-gemini-cli/gemini-2.5-pro
+    gemini: google-gemini-cli/gemini-2.5-flash
+    gpt: openai-codex/gpt-5.2
+capabilities:
+  - text_generation
+  - streaming
+  - function_calling
+  - vision
+  - multi_provider
+  - skill_rewriting
+default_options:
+  temperature: 0.7
+  skills_dir: .pi/skills

data/CHANGELOG.md

@@ -0,0 +1,457 @@
+# 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]
+
+## [0.27.0] - 2026-03-23
+
+### Changed
+- Updated gemspec description to include Gemini CLI, OpenCode, and pi-agent alongside Claude Code and Codex.
+- Removed scaffolding boilerplate comment from gemspec.
+
+## [0.26.2] - 2026-03-23
+
+### Changed
+- Refreshed README to the current package layout pattern with concise sections for how it works, use cases, features, quick start, and ACE linkage.
+
+## [0.26.1] - 2026-03-22
+
+### Changed
+- Refreshed README structure with a dedicated purpose section, ACE-native test/lint command guidance, updated contributing workflow notes, and the canonical ACE footer link.
+
+## [0.26.0] - 2026-03-21
+
+### Changed
+- Added initial `TS-LLMCLI-001` value-gated smoke E2E coverage for `ace-llm-providers-cli-check`, including deterministic no-tools and stubbed-tools command-path validation.
+
+### Technical
+- Added E2E runner/verifier artifacts and decision-record evidence for ADD/SKIP test-case gating.
+
+## [0.25.7] - 2026-03-18
+
+### Technical
+- Added Claude command coverage for explicit tool allowlists so provider presets can distinguish tool-assisted read-only mode from strict no-tools prompt-only mode.
+
+## [0.25.6] - 2026-03-18
+
+### Fixed
+- Preserve explicit empty CLI argument values in array-based `cli_args` normalization so flags like Claude `--tools ""` are forwarded correctly instead of losing their required value.
+
+### Technical
+- Added regression coverage for empty-value CLI arg normalization and Claude command construction.
+
+## [0.25.5] - 2026-03-18
+
+### Fixed
+- Fixed Claude response extraction when the provider payload returned nested JSON content instead of a top-level string, preventing valid outputs from being dropped as empty.
+
+### Technical
+- Added nested payload parsing coverage for `parse_claude_response` to ensure future response-shape changes continue saving valid LLM output.
+
+## [0.25.4] - 2026-03-17
+
+### Fixed
+- Raised `ProviderError` with structured payload details when Claude CLI exits successfully but returns empty response text.
+
+### Technical
+- Added regression coverage for parsed empty-response and `is_error` payload paths in `parse_claude_response`.
+
+## [0.25.3] - 2026-03-17
+
+### Fixed
+- Rescue `Errno::EPIPE` in `SafeCapture` stdin write so broken-pipe crashes (e.g., Codex instant exit) capture stderr for the real error instead of raising an unhandled exception.
+- Detect Claude CLI empty responses (exit 0 but 0-byte output) and raise `ProviderError` instead of silently writing empty files.
+- Removed deprecated `--allowed-tools read_file` from Gemini CLI command builder; `--approval-mode` alone controls tool access.
+
+### Technical
+- Updated Gemini client tests to assert `--allowed-tools` is no longer present in built commands.
+
+## [0.25.2] - 2026-03-15
+
+### Fixed
+- Removed unsupported CLI temperature forwarding for Claude Code CLI adapters so `--temperature` is no longer passed to subprocesses.
+
+## [0.25.1] - 2026-03-12
+
+### Fixed
+- Resolved CLI-provider subprocess working directory from explicit `working_dir`, then `PROJECT_ROOT_PATH`, before falling back to process cwd, so sandbox-relative E2E artifact writes no longer leak into the repo root.
+- Updated Codex, Claude, Claude OAI, Gemini, OpenCode, and Pi clients to use the resolved execution root for subprocess `chdir` and provider-side skill/worktree/prompt-cache path discovery.
+
+## [0.25.0] - 2026-03-12
+
+### Changed
+- Codex now resolves skills from `.codex/skills` only and no longer falls back to `.agent/skills` or `.claude/skills`.
+- Updated the default Codex provider `skills_dir` configuration to `.codex/skills`.
+
+## [0.24.0] - 2026-03-10
+
+### Changed
+- PI now resolves skills from `.pi/skills` as its provider-native directory and no longer falls back to `.agent/skills` or `.claude/skills`.
+- Updated the default PI provider `skills_dir` configuration to `.pi/skills`.
+
+## [0.23.0] - 2026-03-09
+
+### Added
+- Generalized provider skills directory resolution with multi-path fallback for Codex, CodexOAI, ClaudeOAI, and Pi clients
+- Default `skills_dir` configuration for Codex and Pi providers pointing to `.agent/skills`
+
+### Technical
+- Added fallback order tests for `codex_client` and `pi_client` skill directory resolution
+
+## [0.22.2] - 2026-03-09
+
+### Technical
+- Added `SkillNameReader` regression coverage to ensure canonical SKILL frontmatter with nested `skill`/`assign` metadata still yields provider skill names.
+- Added malformed-frontmatter guard test coverage for `SkillNameReader`.
+
+## [0.22.1] - 2026-03-05
+
+### Fixed
+- `SessionFinder` dispatcher test now exercises the actual dispatch path instead of calling the Claude finder directly.
+
+### Changed
+- Added explanatory comment in `OpenCodeSessionFinder` documenting why `project_id` serves as a nil-gate only (OpenCode sessions lack project references).
+- Added explanatory comment in `ClaudeSessionFinder` documenting why substring matching (`include?`) is used instead of exact equality.
+
+## [0.22.0] - 2026-03-05
+
+### Added
+- Provider-specific session finders for Claude, Codex, Pi, Gemini, and OpenCode that detect fork sessions by scanning local session storage and matching by prompt.
+- `SessionFinder` molecule dispatcher that routes session detection to the correct provider-specific atom.
+
+## [0.21.1] - 2026-03-05
+
+### Technical
+- Documented `--output-last-message` minimum version requirement for Codex CLI in README, including verification command and graceful-degradation note.
+
+## [0.21.0] - 2026-03-05
+
+### Added
+- `--output-last-message <path>` flag in `CodexClient#build_codex_command` when `last_message_file:` option is provided, enabling progressive last-message capture for timeout-resilient stall diagnosis.
+
+## [0.20.1] - 2026-03-04
+
+### Changed
+- Gemini CLI prompt staging now uses `.ace-local/llm/prompts`.
+
+
+## [0.20.0] - 2026-03-04
+
+### Added
+- Split shell-style tokens within array CLI arguments so multi-word entries like `"--sandbox danger-full-access"` normalize to discrete CLI flags/values.
+
+## [0.19.4] - 2026-03-04
+
+### Fixed
+- Normalize timeout input in CLI provider capture so numeric and numeric-string values are accepted consistently for dry-run provider execution.
+
+## [0.19.3] - 2026-02-23
+
+### Changed
+- Refactored cli-check script into ATOM structure with ProviderDetector, AuthChecker atoms and HealthChecker molecule
+
+### Technical
+- Updated internal dependency version constraints to current releases
+
+## [0.19.2] - 2026-02-22
+
+### Fixed
+- Document that first matching tier wins when multiple tiers map to the same model in `ClaudeOaiClient`
+
+## [0.19.1] - 2026-02-22
+
+### Fixed
+- `ClaudeOaiClient` now passes tier alias (`sonnet`/`opus`/`haiku`) to `--model` instead of backend model name (`glm-5`), fixing claude CLI model recognition errors
+- Sets `ANTHROPIC_DEFAULT_<TIER>_MODEL` env var so tier alias resolves to backend model name at runtime
+
+### Added
+- `model_tiers` mapping in backend config (`claudeoai.yml`) to associate backend models with Claude CLI tier aliases
+- `resolve_model_tier` method for tier lookup with `sonnet` fallback
+
+## [0.19.0] - 2026-02-22
+
+### Added
+- Git worktree sandbox support for `CodexClient` and `CodexOaiClient` — automatically appends `--add-dir` with the common `.git/` directory when running inside a git worktree, allowing Codex sandbox to write git metadata (index.lock etc.)
+- New `WorktreeDirResolver` atom that detects git worktree environments and resolves the common git directory path
+
+## [0.18.0] - 2026-02-22
+
+### Added
+- New `ClaudeOaiClient` provider for Claude over Anthropic-compatible APIs (Z.ai, OpenRouter)
+- Provider config `claudeoai.yml` with Z.ai backend targeting Anthropic-compatible endpoint (`/api/anthropic`)
+- Backend env injection: sets `ANTHROPIC_BASE_URL`, `ANTHROPIC_AUTH_TOKEN`, and clears `ANTHROPIC_API_KEY` on subprocess
+- Skill command rewriting support via `CommandRewriter` and `SkillNameReader`
+- Tests for backend env injection, command building, model splitting, availability validation, and JSON response parsing
+
+## [0.17.1] - 2026-02-22
+
+### Changed
+- Update codexoai provider: add `name` field to backend config, remove `wire_api` setting, rename env key to `ZAI_API_KEY`
+- `CodexOaiClient` now passes provider `name` to codex via `-c` config override, falls back to backend key when name not configured
+- Remove `wire_api` config override from `CodexOaiClient` (no longer needed)
+
+### Technical
+- Updated tests for new provider name assertion, env key rename, and name fallback behavior
+
+## [0.17.0] - 2026-02-22
+
+### Added
+- New `CodexOaiClient` multi-backend provider that wraps `codex` CLI to target any OpenAI-compatible endpoint via `-c` flag overrides
+- Provider config `codexoai.yml` with Z.ai backend (glm-5, glm-4.7, glm-4.6) and model aliases
+- Tests for command building, backend model splitting, availability validation, and generate flow
+
+### Removed
+- `CodexOSSClient` and `codexoss.yml` provider (called non-existent `codex-oss` binary)
+
+## [0.16.10] - 2026-02-22
+
+### Changed
+- Migrate skill naming and invocation references to hyphenated `ace-*` format (no underscores).
+
+## [0.16.9] - 2026-02-21
+
+### Fixed
+- `ClaudeCodeClient` now merges `subprocess_env` into the subprocess environment, ensuring env vars like `ACE_TMUX_SESSION` are set on the actual `claude -p` process instead of only appearing as text in the prompt
+
+### Added
+- Unit tests for `ClaudeCodeClient` covering env merging, nil handling, and subprocess_env forwarding from `generate()`
+
+## [0.16.8] - 2026-02-21
+
+### Added
+- Process-group lifecycle hardening in `SafeCapture`: isolate subprocesses in dedicated process groups and clean up descendants on timeout and successful parent exit
+- Regression tests for success-path and timeout-path descendant cleanup in `SafeCapture`
+- Debug lifecycle tracing for CLI subprocess management via `ACE_LLM_DEBUG_SUBPROCESS=1`
+
+### Changed
+- `ClaudeCodeClient` now emits debug subprocess context (timeout, command, prompt size) when `ACE_LLM_DEBUG_SUBPROCESS=1`
+
+## [0.16.7] - 2026-02-21
+
+### Changed
+- Update skill name handling and documentation for colon-free convention (`ace_domain_action` format)
+- Update command rewriter tests for new skill naming convention
+
+## [0.16.6] - 2026-02-17
+
+### Fixed
+- Added `SafeCapture` molecule test coverage to verify `env:` parameter propagation into subprocess execution (`test_env_option_passed_to_subprocess`)
+
+## [0.16.5] - 2026-02-15
+
+### Changed
+- **ClaudeCodeClient**: Optimize subprocess environment — pass minimal env override `{"CLAUDECODE" => nil}` instead of copying entire `ENV.to_h`
+
+## [0.16.4] - 2026-02-15
+
+### Fixed
+- **ClaudeCodeClient**: Pass prompt via stdin instead of CLI argument to avoid Linux `MAX_ARG_STRLEN` (128KB) limit on large prompts
+- **ClaudeCodeClient**: Remove `--system-prompt` and `--append-system-prompt` CLI args (system content already embedded in formatted prompt)
+
+### Changed
+- **CodexClient**: `--sandbox` mode is now caller-controlled via `options[:sandbox]` instead of hardcoded `read-only`
+
+## [0.16.3] - 2026-02-15
+
+### Fixed
+- **ClaudeCodeClient**: Clear `CLAUDECODE` env var before spawning subprocess to allow `claude -p` (one-shot mode) to run from within a Claude Code session — works around nested session guard added in Claude Code v2.1.41
+- **CodexClient**: Add `--sandbox read-only` to prevent agent from executing commands instead of reviewing the prompt content
+- **SafeCapture**: Added optional `env:` parameter for subprocess environment control
+
+## [0.16.2] - 2026-02-07
+
+### Fixed
+- **PiClient**: Thread-safety fix — removed `@pending_system_prompt` shared mutable state; `build_full_prompt` now returns `[prompt, system_prompt]` tuple passed explicitly to `build_pi_command`
+- **SkillCommandRewriter**: Replaced misleading "Deprecated" docstring with accurate "Convenience wrapper" description
+
+### Changed
+- **SafeCapture**: Moved from `atoms/` to `molecules/` to comply with ATOM architecture (performs I/O via Open3, Process.kill, Thread spawning)
+- Updated all 6 CLI clients to reference `Molecules::SafeCapture` instead of `Atoms::SafeCapture`
+
+### Technical
+- 215 tests, 511 assertions, 0 failures (10 skipped)
+
+## [0.16.1] - 2026-02-07
+
+### Fixed
+- **PiClient**: Handle nested provider pattern with colon (e.g., `openrouter:openai/gpt-oss-120b`)
+  - Updated `split_provider_model` to correctly parse provider:model format
+  - Now splits on `:` before `/` when colon is present in model string
+  - Added regression tests for standard provider/model format
+
+### Technical
+- 2 new tests for nested provider model parsing
+
+## [0.16.0] - 2026-02-07
+
+### Added
+- **CodexClient**: Skill command rewriting support transforms `/name` → `$name` for Codex CLI skill discovery
+- **CodexClient**: `skill_rewriting` capability and `skills_dir` configuration option in provider defaults
+
+### Changed
+- Refactored skill command rewriting into provider-agnostic `CommandRewriter` base class with configurable formatter proc
+- `SkillCommandRewriter` now a thin wrapper using `CommandRewriter` with `PI_FORMATTER` for backward compatibility
+- `PiClient` updated to use `CommandRewriter` with `CommandFormatters::PI_FORMATTER`
+
+### Technical
+- New `CommandRewriter` atom in `atoms/command_rewriter.rb` for provider-agnostic command rewriting
+- New `CommandFormatters` module in `atoms/command_formatters.rb` with `PI_FORMATTER` and `CODEX_FORMATTER`
+- 20 tests for `CommandRewriter` covering both Pi and Codex formatters
+- 25 tests for `CodexClient` covering skill rewriting and configuration
+
+## [0.15.0] - 2026-02-07
+
+### Added
+- **PiClient**: New CLI provider for Pi (v0.52.7), a multi-provider terminal AI agent
+  - Supports ZAI, Anthropic, Google Gemini, and OpenAI Codex models via Pi's unified platform
+  - Skill command rewriting: transforms `/name` → `/skill:name` for Pi's skill discovery
+  - NDJSON response parsing with plain text fallback
+  - System prompt support via native `--system-prompt` flag
+  - 17 tests for `SkillCommandRewriter` atom covering code blocks, URLs, edge cases
+  - 37 tests for `PiClient` covering command building, prompt preprocessing, NDJSON parsing
+
+### Technical
+- New `SkillCommandRewriter` atom in `atoms/skill_command_rewriter.rb` for pure function skill command rewriting
+- New `SkillNameReader` molecule in `molecules/skill_name_reader.rb` for reading skill names from SKILL.md frontmatter
+
+## [0.14.2] - 2026-02-06
+
+### Fixed
+- Replace unsafe `Timeout.timeout { Open3.capture3 }` with thread-safe `SafeCapture` atom using `Open3.popen3` + `Process.kill` for process-level timeout — eliminates "stream closed in another thread (IOError)" errors during parallel CLI execution
+
+### Technical
+- New `SafeCapture` atom in `atoms/safe_capture.rb` with 7 tests covering stdout/stderr capture, timeout kill, stdin_data, chdir, and error messages
+- Updated all 5 CLI clients (Claude, Codex, Codex OSS, Gemini, OpenCode) to use `SafeCapture`
+- Updated Gemini client test stub from `Open3.capture3` to `SafeCapture.call`
+
+## [0.14.1] - 2026-02-06
+
+### Fixed
+- Honor `--` sentinel in ArgsNormalizer: args after `--` now pass through verbatim without auto-prefixing
+- User CLI args now take precedence over command-generated flags (moved to end of command construction in all 4 clients)
+
+## [0.14.0] - 2026-02-05
+
+### Added
+- CLI argument passthrough support with normalization for CLI providers
+
+### Fixed
+- Improved CLI argument error reporting and provider command wiring
+
+## [0.13.2] - 2026-01-22
+
+### Fixed
+- **OpenCodeClient**: Prevent client hang on 400 error and improve output
+  - Added `stdin_data: ""` to `Open3.capture3` to prevent hanging on interactive prompts
+  - Added `--format json` flag for structured output
+  - Improved 400 Bad Request error detection with clearer error messages
+  - Updated test expectations for `--format json` flag
+
+## [0.13.1] - 2026-01-16
+
+### Fixed
+- **OpenCodeClient**: Fixed command syntax to match current OpenCode CLI interface
+  - Changed from `opencode generate` to `opencode run` subcommand
+  - Pass prompt as positional argument instead of `--prompt` flag
+  - Removed unsupported flags: `--format`, `--temperature`, `--max-tokens`, `--system`
+  - Handle system prompts by prepending to main prompt (no native `--system` flag)
+- Added regression tests for correct OpenCode command building
+
+## [0.13.0] - 2026-01-13
+
+### Added
+- **GeminiClient**: Added Google Gemini CLI provider integration
+  - Supports Gemini 2.5 Flash, Gemini 2.5 Pro, Gemini 2.0 Flash, and Gemini 1.5 Pro models
+  - JSON output parsing for structured responses with token metadata
+  - System prompt embedding (Gemini CLI lacks native `--system-prompt` flag)
+  - Provider aliases: `gflash`, `gpro`, `gemini-flash`, `gemini-pro`
+  - Auto-registers with ace-llm provider system
+
+## [0.12.0] - 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
+
+## [0.11.1] - 2025-12-30
+
+### Changed
+
+* Replace ace-support-core dependency with ace-config
+
+## [0.11.0] - 2025-12-30
+
+### Changed
+
+* Rename `.ace.example/` to `.ace-defaults/` for gem defaults directory
+
+
+## [0.10.2] - 2025-12-06
+
+### Technical
+- Updated ace-llm dependency from `~> 0.12.0` to `~> 0.13.0` for OpenRouter provider support
+
+## [0.10.1] - 2025-11-17
+
+### Technical
+- Updated ace-llm dependency from `~> 0.10.0` to `~> 0.11.0` for graceful provider fallback support
+
+## [0.10.0] - 2025-11-15
+
+### Added
+- **ClaudeCodeClient Enhancement**: Added support for `--append-system-prompt` flag
+  - Maps `system_append` option to Claude CLI's `--append-system-prompt` flag
+  - Enables flexible prompt composition with Claude models
+
+### Fixed
+- **ClaudeCodeClient Bug**: Fixed system prompt handling to use correct `--system-prompt` flag
+  - Changed from non-existent `--system` to proper `--system-prompt` flag
+  - Resolves issue where system prompts were ignored with Claude provider
+
+### Changed
+- **Dependency Update**: Updated ace-llm dependency to ~> 0.10.0
+  - Aligns with ace-llm minor version bump for system prompt control features
+
+### Technical
+- Added deprecation note for `append_system_prompt` option, prefer `system_append` for consistency
+
+## [0.9.3] - 2025-10-08
+
+### Changed
+
+- **Test Structure Reorganization**: Reorganized tests with proper ATOM categorization
+  - Moved `test/ace/llm_providers_cli_test.rb` → `test/llm_providers_cli_test.rb`
+  - Created `test/molecules/` for CLI provider tests
+  - Created `test/edge/` for edge case tests (new pattern)
+  - Created `test/integration/` for provider registration tests
+  - Fixed require paths to match new structure
+  - Aligns with standardized flat ATOM structure across all ACE packages
+
+## [0.9.2] - 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
+
+## [0.9.1] - 2025-10-07
+
+### Fixed
+- Standardized test file naming to follow project convention (`*_test.rb` suffix instead of `test_*` prefix)
+- Renamed `test_cli_execution_edge.rb` → `cli_execution_edge_test.rb`
+- Renamed `test_cli_providers.rb` → `cli_providers_test.rb`
+- Renamed `test_provider_registration.rb` → `provider_registration_test.rb`
+- Test count increased from 51 to 57 tests (all tests now properly discovered)
+
+## [0.9.0] - 2024-XX-XX
+
+### Added
+- Initial release with CLI-based LLM provider support
+- Support for Claude Code and Codex providers
+- Provider registration and configuration system
+- OpenCode client implementation

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.