Foundry MCP

# 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/1.1.0/), and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html). ## [Unreleased] ## [0.9.0b12] - 2026-02-01 ### Added - **Claude ZAI Provider**: New CLI provider for users with custom `claude-zai` alias configurations - Separate provider from `claude` for custom alias settings (e.g., custom MCP servers, system prompts) - Read-only tool restrictions for security: allows Read, Grep, Glob, read-only git commands - Blocks write operations, web operations (data exfiltration risk), and destructive commands - Detector registration with `FOUNDRY_CLAUDE_ZAI_AVAILABLE_OVERRIDE` and `FOUNDRY_CLAUDE_ZAI_BINARY` env vars - Sample config updated with claude-zai provider examples ## [0.9.0b11] - 2026-01-28 ### Added - **Document Digest for Deep Research**: Integrated document digest pipeline into deep research workflows - PDF extraction with security hardening (file size limits, page limits, malformed PDF handling) - Foundation models and configuration for document digest processing - Deep research integration for digest-based source analysis - Comprehensive test coverage for document digest, deep research digest integration, and PDF extraction ### Changed - Updated MCP response schema documentation with expanded contract details ## [0.9.0b10] - 2026-01-27 ### Added - **Provider Rate Limit Resilience**: Unified resilience layer for all search providers with circuit breakers, rate limiting, and retry with backoff - **Circuit Breakers**: Per-provider circuit breakers with CLOSED → OPEN → HALF_OPEN state transitions - Opens after 5 consecutive failures, 30s recovery timeout - HALF_OPEN allows recovery probes before fully closing - Deep research skips providers with OPEN circuits, enabling graceful failover - **Rate Limiting**: Token bucket rate limiting per provider - Default: 1 RPS with burst limit of 3 - Semantic Scholar: 0.9 RPS with burst limit of 2 (conservative for academic API) - Configurable `max_wait_seconds` for rate limit queue timeout - **Retry with Jitter**: Exponential backoff with deterministic jitter (seeded RNG for testing) - 50-150% jitter range prevents thundering herd - Respects time budget constraints during retry delays - **Error Classification**: Provider-specific error handling via `classify_error()` hook - 429 errors: Retryable, does NOT trip circuit breaker - 401/403 errors: Not retryable, does NOT trip circuit breaker - 5xx errors: Retryable, trips circuit breaker - Timeouts/network errors: Retryable, trips circuit breaker - **Time Budget Enforcement**: Operations cancelled when exceeding time budget - **Graceful Degradation**: Providers that trip mid-gathering are skipped for remaining sub-queries - **Observability**: Circuit breaker state changes and rate limit waits logged via audit_log - **Testing Support**: `reset_resilience_manager_for_testing()` for test isolation - **New Exceptions**: - `RateLimitWaitError`: Raised when rate limit wait would exceed `max_wait_seconds` - `TimeBudgetExceededError`: Raised when operation exceeds time budget ### Changed - Deep research gathering phase now filters providers by circuit breaker state before execution - Provider docstrings updated with resilience configuration details ## [0.9.0b9] - 2026-01-27 ### Added - **Semantic Scholar API Enhancements**: Extended Semantic Scholar search provider with TLDR support and new filtering/sorting capabilities - **TLDR Summaries**: Auto-generated paper summaries now used as snippet when available, with abstract fallback - **Extended Metadata**: New fields in search results: `venue`, `influential_citation_count`, `reference_count`, `fields_of_study`, `tldr` - **Publication Type Filtering**: `publication_types` parameter to filter by paper type (JournalArticle, Conference, Review, etc.) - **Sorting**: `sort_by` parameter (citationCount, publicationDate, paperId) with `sort_order` (asc/desc, default: desc) - **Extended Fields Toggle**: `use_extended_fields` parameter (default: True) to control metadata verbosity - **Parameter Validation**: Comprehensive validation with clear error messages for all new parameters - **Configuration Fields**: New `[research]` TOML config options: - `semantic_scholar_publication_types`, `semantic_scholar_sort_by`, `semantic_scholar_sort_order` - `semantic_scholar_use_extended_fields` - **Deep Research Integration**: Parameters automatically propagated from config to Semantic Scholar search calls ### Changed - **Semantic Scholar Endpoint**: Switched from `/paper/search/bulk` to `/paper/search` (relevance search) to enable TLDR support - **Breaking**: Maximum results per query reduced from 1000 to 100 (API limit for /paper/search endpoint) - Results are now relevance-ranked by default - **Semantic Scholar Sorting**: `sort_order` without `sort_by` now defaults to `publicationDate` ## [0.9.0b8] - 2026-01-27 ### Added - **Perplexity API Enhancements**: Extended Perplexity search provider with all parameters from the Perplexity Sonar API spec - **Search Context Size**: `search_context_size` parameter (`"low"`, `"medium"`, `"high"`) controls result context depth - **Token Limits**: `max_tokens` (default: 50000) and `max_tokens_per_page` (default: 2048) for response control - **Recency Filtering**: `recency_filter` parameter (`"day"`, `"week"`, `"month"`, `"year"`) for time-based filtering - **Date Range Filters**: `search_after_date` and `search_before_date` for precise date range queries (MM/DD/YYYY format) - **Content Modification Filters**: `last_updated_after_filter` and `last_updated_before_filter` for filtering by modification date - **Geographic Filtering**: `country` parameter (ISO 3166-1 alpha-2 code, e.g., `"US"`) - **Configuration Fields**: New `[research]` TOML config options: - `perplexity_search_context_size`, `perplexity_max_tokens`, `perplexity_max_tokens_per_page` - `perplexity_recency_filter`, `perplexity_country` - **Deep Research Integration**: Parameters automatically propagated from config to Perplexity search calls - **Parameter Validation**: Comprehensive validation with clear error messages for all new parameters ### Migration **Default behavior unchanged**: All new config fields have defaults that preserve existing behavior. Users not configuring Perplexity options will see no change. ### Fixed - **Environment get-config Lookup**: Fixed `environment(action="get-config")` to use the full config lookup hierarchy (matching `ServerConfig.from_env()`), instead of only checking the current working directory - Added `path` parameter for explicit config file specification - Now searches: explicit path → `FOUNDRY_MCP_CONFIG_FILE` env var → project dir → `~/.foundry-mcp.toml` → XDG config - Error response includes `searched_paths` for debugging ## [0.9.0b7] - 2026-01-26 ### Added - **Deep Research CPU Optimization**: Performance improvements to reduce CPU and I/O overhead in deep research workflows - **Status Persistence Throttling**: New `status_persistence_throttle_seconds` config (default: 5) limits disk writes during research, persisting immediately only on phase/iteration changes or terminal states - **Token Count Caching**: Content hash-based caching in `ResearchSource` eliminates redundant token estimation calls on workflow resume - **Audit Verbosity Modes**: New `audit_verbosity` config option (`"full"` or `"minimal"`) controls JSONL audit payload size - minimal mode nulls large text fields while preserving schema shape for downstream compatibility ### Changed - Status persistence now uses smart throttling instead of persisting after every operation - Token estimation skips already-cached sources, reducing CPU on resume operations ## [0.9.0b6] - 2026-01-26 ### Added - **Deep Research Resilience**: Comprehensive resilience improvements for long-running research workflows - **Task Registry & Cancellation**: Centralized task registry with two-phase cancellation (cooperative then forced) - **Timeout Watchdog**: Background monitor that detects timeout and staleness conditions - **Provider Timeout Enforcement**: Per-provider timeout handling with `ProviderTimeoutError` - **Progress Heartbeats**: `last_heartbeat_at` timestamp updated before each provider call for visibility - **Default Timeout Configuration**: Configurable `deep_research_timeout` with precedence rules (explicit > config > 600s fallback) - **Executor Isolation**: `ProviderExecutor` thread pool for isolating blocking CLI provider operations - **Status Metadata**: New fields in status response: `last_heartbeat_at`, `is_timed_out`, `is_stale`, `effective_timeout` - **Documentation**: Added resilience configuration and troubleshooting guides - Timeout behavior and configuration in `docs/06-configuration.md` - Troubleshooting for timeout, staleness, and cancellation issues in `docs/07-troubleshooting.md` ## [0.9.0b5] - 2026-01-26 ### Fixed - **TOCTOU Race Conditions**: Fixed time-of-check-time-of-use race conditions in `FileStorageBackend` by moving file existence and expiry checks inside file locks - **Thread Safety**: Added `_active_sessions_lock` for thread-safe session tracking in deep research, preventing race conditions during concurrent access - **Task Registry GC**: Replaced `WeakValueDictionary` with regular dict for background task registry to prevent premature garbage collection of running tasks - **Timeout Flag Accuracy**: Improved timeout metadata accuracy - only sets `timeout: true` when all failures were timeouts, not when mixed with other error types - **State Persistence**: Save thread/consensus state BEFORE provider calls to ensure user messages and responses persist even if provider fails - **Deep Research Timeout Handling**: Mark deep research state as failed on timeout instead of leaving in intermediate state - **Orphaned Lock Files**: Clean up orphaned lock files when data files are missing ### Changed - Use timezone-aware `datetime.now(timezone.utc)` throughout instead of deprecated `datetime.utcnow()` - Added task cleanup methods (`cleanup_stale_tasks`, `_cleanup_completed_task`) for memory management in long-running processes ## [0.9.0b4] - 2026-01-25 ### Added - **Tavily API Enhancements**: Extended Tavily search provider with all parameters from the Tavily API spec and added new Extract endpoint support - **New Search Parameters**: - `search_depth`: Search mode - `"basic"` (default), `"advanced"` (2x credits), `"fast"`, `"ultra_fast"` - `topic`: Search topic - `"general"` (default), `"news"` - `days`: News recency limit (1-365 days, only when `topic="news"`) - `include_images`: Include image results (default: false) - `include_raw_content`: Get raw page content - `false`, `true`/`"markdown"`, or `"text"` - `country`: ISO 3166-1 alpha-2 country code to boost results (e.g., `"US"`) - `chunks_per_source`: Chunk count for advanced search (1-5, default: 3) - `auto_parameters`: Let Tavily auto-configure based on query intent - **Tavily Extract Provider** (`TavilyExtractProvider`): New provider for URL content extraction - Extracts structured content from up to 10 URLs per request - Parameters: `extract_depth` (`"basic"`/`"advanced"`), `include_images`, `format` (`"markdown"`/`"text"`), `query`, `chunks_per_source` - SSRF protection: blocks private IPs, localhost, reserved ranges, suspicious schemes - Partial failure handling: returns successful extractions even when some URLs fail - Retry logic with exponential backoff matching search provider - **Configuration Fields**: New `[research]` TOML config options - `tavily_search_depth`, `tavily_topic`, `tavily_news_days`, `tavily_include_images` - `tavily_country`, `tavily_chunks_per_source`, `tavily_auto_parameters` - `tavily_extract_depth`, `tavily_extract_include_images` - `tavily_extract_in_deep_research`: Enable extract as follow-up step in deep research - `tavily_extract_max_urls`: Max URLs to extract per deep research run (default: 5) - **Deep Research Integration**: - `_get_tavily_search_kwargs()` method propagates config to search calls - Research mode smart defaults: academic/technical modes prefer `"advanced"` depth - Optional extract follow-up step when `tavily_extract_in_deep_research=true` - **Security**: URL validation for extract with SSRF protection - Blocks localhost, private IPs (10.x, 172.16.x, 192.168.x), link-local, loopback - Blocks dangerous schemes (file://, gopher://, dict://, data://) - DNS resolution with timeout to prevent DNS rebinding - Max URL length (2048 chars), max content size (50KB per source) ### Migration **Default behavior unchanged**: All new config fields have defaults that preserve existing behavior. Users not configuring Tavily options will see no change. **Credit cost awareness**: Using `search_depth="advanced"` doubles Tavily API credit usage. Consider `"basic"` for most queries, `"advanced"` only when deeper analysis is needed. **Configuration example**: ```toml [research] # Search configuration tavily_search_depth = "basic" # or "advanced" (2x credits), "fast", "ultra_fast" tavily_topic = "general" # or "news" tavily_news_days = 7 # only when topic = "news" tavily_include_images = false tavily_country = "US" # boost results from country tavily_chunks_per_source = 3 # 1-5, for advanced search tavily_auto_parameters = false # let Tavily auto-configure # Extract configuration tavily_extract_depth = "basic" # or "advanced" tavily_extract_include_images = false tavily_extract_in_deep_research = false # enable extract follow-up tavily_extract_max_urls = 5 # max URLs per deep research run ``` ## [0.9.0b3] - 2026-01-25 ### Added - **XDG Base Directory support**: Config loading now checks `~/.config/foundry-mcp/config.toml` (or `$XDG_CONFIG_HOME/foundry-mcp/config.toml`) as the lowest-priority user config location, following the XDG Base Directory Specification common on Linux systems ### Changed - **Config loading priority** (highest to lowest): 1. Environment variables 2. Project config (`./foundry-mcp.toml`) 3. User config (`~/.foundry-mcp.toml`) 4. XDG config (`~/.config/foundry-mcp/config.toml`) - NEW 5. Built-in defaults ## [0.9.0b2] - 2026-01-25 ### Fixed - **Deep research timezone crash**: Fixed `TypeError: can't subtract offset-naive and offset-aware datetimes` in `_allocate_source_budget` when Tavily returns timezone-naive `discovered_at` timestamps ## [0.9.0b1] - 2026-01-25 ### Added - **Deep research token management**: Intelligent content compression and graceful degradation for large research workflows - New `ContextBudgetAllocator` manages token budget allocation with priority-based content selection - New `ContentArchive` enables file-based storage of dropped content with TTL cleanup - New `GracefulDegradationManager` applies progressive content compression (FULL → CONDENSED → COMPRESSED → KEY_POINTS → HEADLINE → DROPPED) - Fidelity tracking with per-item compression ratios and overall fidelity score - Top-5 source protection guardrail ensures minimum 30% fidelity for high-priority items - Content summarization via configurable LLM providers with fallback chain - **Token management configuration** (`ResearchConfig`): - `token_management_enabled` - Master switch for token management (default: true) - `token_safety_margin` - Budget safety buffer fraction (default: 0.15) - `runtime_overhead` - Tokens reserved for CLI/IDE context (default: 60000 for Claude Code) - `model_context_overrides` - Per-model context/output limit overrides - `summarization_provider` / `summarization_providers` - LLM providers for content summarization - `summarization_timeout` / `summarization_cache_enabled` - Summarization performance settings - `allow_content_dropping` - Allow dropping low-priority content (default: false) - `content_archive_enabled` / `content_archive_ttl_hours` / `research_archive_dir` - Archive settings - **Response schema extensions** for content fidelity: - `meta.content_fidelity` - Response completeness level (full|partial|summary|reference_only) - `meta.content_fidelity_schema_version` - Schema version for fidelity metadata ("1.0") - `meta.dropped_content_ids` - IDs of content items omitted from response - `meta.content_archive_hashes` - Map of archive IDs to content hashes for retrieval - `meta.warning_details` - Structured warnings with code, severity, message, and context - **Token management warning codes**: - `CONTENT_TRUNCATED` - Content summarized/compressed to fit budget - `CONTENT_DROPPED` - Low-priority content removed - `TOKEN_BUDGET_FLOORED` - Item preserved due to min items guardrail - `ARCHIVE_WRITE_FAILED` / `ARCHIVE_DISABLED` / `ARCHIVE_READ_CORRUPT` - Archive status - `TOKEN_COUNT_ESTIMATE_USED` - Character-based heuristic fallback (tiktoken unavailable) - **Documentation**: - Token Management section in `docs/concepts/deep_research_workflow.md` - Token Management Warnings section (17.4) in `dev_docs/codebase_standards/cli-output.md` - Sample TOML configuration with runtime overhead values for different CLIs ### Changed - **`success_response` helper** now accepts content fidelity parameters (`content_fidelity`, `dropped_content_ids`, `content_archive_hashes`, `warning_details`) ## [0.8.35] - 2026-01-24 ### Added - **Deep research polling mitigation**: Track status check counts and provide guidance to prevent aggressive polling - New `status_check_count` and `last_status_check_at` fields in `DeepResearchState` - Start response now includes `polling_guidance` with max checks, typical duration, and behavioral rules - Status response includes `next_action` guidance based on check count (warns at limit) - Instructs callers not to use WebSearch/WebFetch while deep research is running ## [0.8.34] - 2026-01-24 ### Added - **Deep research timeout resilience**: Async provider execution with timeout protection and retry/fallback logic - New `_execute_provider_async()` method in base workflow with `asyncio.wait_for()` timeout protection - All 4 deep research phase methods now use async execution with timeout handling - Per-phase fallback provider lists: `deep_research_planning_providers`, `deep_research_analysis_providers`, `deep_research_synthesis_providers`, `deep_research_refinement_providers` - Retry settings: `deep_research_max_retries` (default: 2) and `deep_research_retry_delay` (default: 5.0s) - `get_phase_fallback_providers()` helper method in `ResearchConfig` - Enhanced timeout logging with phase context and provider history - Sample TOML documentation for new retry/fallback settings ## [0.8.33] - 2026-01-24 ### Added - **Research workflow exception handling**: Improved resilience to prevent MCP server crashes - `DeepResearchWorkflow.execute()` now catches all exceptions and returns error result - `_dispatch_research_action()` catches exceptions and returns structured error response - Added warning logging when provider resolution fails in base workflow - Added comprehensive tests for exception handling behavior ## [0.8.32] - 2026-01-22 ### Changed - **Improved lifecycle action descriptions**: More descriptive summaries for `activate`, `complete`, and `archive` actions clarifying behavior and folder destinations ### Removed - **Removed intake tools feature flag**: Intake tools (`intake-add`, `intake-list`, `intake-dismiss`) are now always enabled - Removed `intake_tools` feature flag from capabilities manifest - Removed feature flag check code from authoring tool handlers - Removed feature flag documentation from intake guide ## [0.8.31] - 2026-01-22 ### Added - **Batch mode for autonomous sessions**: New workflow mode that executes a fixed number of tasks before pausing for approval - `AutonomousSession` now tracks `batch_mode`, `batch_size`, and `batch_remaining` - Task `complete` action decrements batch counter and auto-pauses when limit reached - Session config `session-config` applies `WorkflowConfig` defaults for batch settings - Supports `WorkflowMode.BATCH` and `WorkflowMode.AUTONOMOUS` from config ### Changed - **Research state storage relocation**: Default storage moved from `~/.foundry-mcp/research` to project-local `specs/.research` - New `research_dir` config option replaces old global storage path - New `FOUNDRY_MCP_RESEARCH_DIR` environment variable - **Sample config updates**: Increased default research timeouts from 360s to 600s, updated provider examples - **Documentation updates**: Various fixes across guides and tool reference ### Removed - **Simplified ResearchConfig**: Removed `storage_path` and `storage_backend` fields (storage is now always project-local) - **Removed `journals_path`**: Replaced with `research_dir` for research state storage - **Removed root config file**: Deleted `foundry-mcp.toml` from repository root (use `samples/foundry-mcp.toml` as template) - **Removed home config sample**: Deleted `samples/home-foundry-mcp.toml` (layered config still works but sample removed) ## [0.8.30] - 2026-01-22 ### Added - **Layered configuration loading**: Configuration now loads in layers for better separation of concerns - Home config (`~/.foundry-mcp.toml`) for user-wide defaults (API keys, preferred providers, logging preferences) - Project config (`./foundry-mcp.toml`) for project-specific overrides - Environment variables for runtime overrides (highest priority) - Legacy `.foundry-mcp.toml` fallback maintained for backwards compatibility - **User config sample**: New `samples/home-foundry-mcp.toml` example for user-level configuration - **Config hierarchy tests**: Comprehensive test coverage for layered configuration loading ### Changed - **Updated configuration documentation**: `docs/06-configuration.md` explains the new config hierarchy with table of file locations and use cases - **Updated project config sample**: `samples/foundry-mcp.toml` updated with layered config priority comments ## [0.8.29] - 2026-01-19 ### Changed - **Renamed bikelane to notes**: Intake queue terminology updated across tools, docs, and specs - Config key `notes_dir` replaces `bikelane_dir` (no backward compatibility) - Env var `FOUNDRY_MCP_NOTES_DIR` replaces `FOUNDRY_MCP_BIKELANE_DIR` - Default storage path is now `specs/.notes/` ## [0.8.28] - 2026-01-18 ### Changed - **Perplexity provider enhancements**: Improved deep research content extraction - Added `include_raw_content` option to map snippets to content field for source aggregation - Added `max_tokens` (50000) and `max_tokens_per_page` (2048) parameters for better content retrieval ### Removed - **Google search provider**: Removed Google Custom Search from deep research providers - Removed `GOOGLE_API_KEY` and `GOOGLE_CSE_ID` environment variables from documentation - Removed `google` from default `deep_research_providers` list - Removed Google rate limit configuration ## [0.8.27] - 2026-01-18 ### Fixed - **Restored sample config sections**: Added back `[workflow]` and `[consultation]` sections to `samples/foundry-mcp.toml` that are required by `llm_config.py` and `ai_consultation.py` ### Removed - **Removed `[metrics_persistence]` from sample config**: Advanced feature not needed in sample configuration ## [0.8.26] - 2026-01-18 ### Changed - **Documentation reorganization**: Separated internal developer docs from user-facing documentation - Moved MCP best practices, CLI best practices, codebase standards, and developer guides to `dev_docs/` - Created new user-facing documentation structure in `docs/` with quick start, workflow guides, CLI/MCP references, configuration, and troubleshooting - Updated `AGENTS.md` and `CLAUDE.md` with corrected paths to `dev_docs/` - **README overhaul**: Rewrote README to be user-focused with clearer value proposition, streamlined installation, and practical quick start guide ### Added - **New user documentation**: Comprehensive docs for end users - `docs/01-quick-start.md` - Getting started guide - `docs/02-core-concepts.md` - Specs, phases, tasks explained - `docs/03-workflow-guide.md` - Step-by-step workflow - `docs/04-cli-command-reference.md` - Full CLI documentation - `docs/05-mcp-tool-reference.md` - MCP tool documentation - `docs/06-configuration.md` - Configuration options - `docs/07-troubleshooting.md` - Common issues and solutions - `docs/concepts/spec-schema.md` - Spec JSON schema reference - `docs/concepts/response-envelope.md` - Response format reference - `docs/reference/error-codes.md` - Error code reference ## [0.8.25] - 2026-01-18 ### Removed - **Dashboard feature**: Removed entire Streamlit-based dashboard package (`src/foundry_mcp/dashboard/`) - **Dashboard CLI command**: Removed `dashboard` subcommand (`src/foundry_mcp/cli/commands/dashboard.py`) - **Dashboard dependencies**: Removed streamlit, plotly, pandas from pyproject.toml - **Dashboard documentation**: Removed `docs/dashboards/` directory and Prometheus rules - **Metrics unified tool**: Removed `src/foundry_mcp/tools/unified/metrics.py` (tool count 17→16) - **Legacy verification type mapping**: Removed `"test"→"run-tests"` and `"auto"→"run-tests"` auto-conversion - **Legacy `phase_id` parameter aliasing**: Removed `phase_id` as alias for `parent` in task actions - **Legacy metadata category fallback**: Removed category fallback logic from metadata validation ### Changed - Updated branding from "legacy claude-sdd-toolkit CLI" to "Foundry CLI" - Simplified `fix-verification-types` action (sets unknown types to `"manual"` only, no legacy mapping) - Updated tool count in documentation (17→16) ## [0.8.24] - 2026-01-17 ### Changed - **Updated skill references for claude-foundry v1.6.8**: Renamed all skill references to match the new unified naming convention - `sdd-toolkit:sdd-pr` → `foundry:foundry-pr` - `sdd-toolkit:sdd-modify` → `foundry:foundry-spec` - `sdd-toolkit:sdd-plan-review` → `foundry:foundry-spec` - Updated 15 occurrences across 4 files: `tools/unified/pr.py`, `tools/unified/review.py`, `core/review.py`, `cli/commands/review.py` - Plugin name changed from `sdd-toolkit` to `foundry` ## [0.8.23] - 2026-01-11 ### Fixed - **Deep research per-phase timeout config parsing**: Fixed config parsing to use correct timeout defaults - Main timeout: 120s → 600s - Planning timeout: 60s → 360s - Analysis timeout: 90s → 360s - Synthesis timeout: 180s → 600s - Refinement timeout: 60s → 360s - Config parsing now matches the class default values ### Enhanced - **Bulk phase task validation for all task types**: Extended `authoring(action="phase-add-bulk")` validation - Now supports all task types: task, subtask, verify, research (previously only task/verify) - Uses `TASK_TYPES` constant from task module for consistency - Added validation for research-specific parameters: `blocking_mode`, `research_type`, `query` - Improved error messages to show all valid task types ## [0.8.22] - 2026-01-07 ### Changed - **Research default timeout increased**: Changed `default_timeout` from 60s to 360s in research config - Provides more time for complex research operations that involve multiple LLM calls and web fetches ### Fixed - **Flexible assumption types**: Removed strict validation on `assumption_type` parameter in assumption operations - Previously only accepted "constraint" or "requirement" - Now accepts any string value (e.g., "architectural", "security", "performance") - Affects `authoring(action="assumption-add")` and `authoring(action="assumption-list")` ## [0.8.21] - 2026-01-07 ### Added - **Review tool "spec-review" alias**: Added `"spec-review"` as an alias for the `"spec"` action in the review tool - Users can now call `review(action="spec-review", ...)` or `review(action="spec", ...)` interchangeably - **MCP task tool batch parameters**: Added missing parameters to task tool MCP schema - `completions: List[Dict]` for `complete-batch` action with per-task completion notes - `threshold_hours: float` for `reset-batch` action to filter tasks by estimated hours ## [0.8.20] - 2026-01-06 ### Added - **Hours hierarchy recalculation actions**: New spec actions to aggregate hours from tasks up through hierarchy - `spec(action="recalculate-hours")`: Sums `estimated_hours` from task/subtask/verify nodes to phases, then to spec total - `spec(action="recalculate-actual-hours")`: Sums `actual_hours` from task/subtask/verify nodes to phases, then to spec total - Both support `dry_run` mode to preview changes without saving - Returns detailed breakdown per phase with previous/calculated/delta values - **Auto-calculate actual_hours on task completion**: When completing tasks with `started_at` but no manual `actual_hours` - Calculates elapsed time from `started_at` to `completed_at` in hours - Applies to both single task completion via `update_task_status()` and batch completion via `complete_batch()` - Preserves manually set values - only auto-calculates if `actual_hours` not already present ## [0.8.19] - 2026-01-06 ### Fixed - **Deep research workflow failure state handling**: Fixed issue where failed research sessions appeared "stalled" instead of properly marked as failed - Added `mark_failed(error: str)` method to `DeepResearchState` to explicitly mark sessions as failed - `_execute_gathering_async` now returns descriptive error message when all sub-queries fail to find sources - All phase failure blocks (planning, gathering, analysis, synthesis) now call `state.mark_failed()` before returning - Status reporting now includes `is_failed` and `failure_error` fields - Status display shows "Failed" instead of "In Progress" for failed sessions with error details ## [0.8.18] - 2026-01-06 ### Fixed - **Deep research `continue`/`resume` event loop conflict**: Fixed `RuntimeError: Cannot run the event loop while another loop is running` - `_continue_research()` now uses same event loop handling pattern as `_start_research()` - Detects running event loop with `asyncio.get_event_loop().is_running()` - Uses `ThreadPoolExecutor` to run async code in separate thread when already in async context - Fixes issue when MCP server calls deep research continue/resume actions ## [0.8.17] - 2026-01-06 ### Fixed - **MCP task tool missing `task_ids` parameter**: Added `task_ids: Optional[List[str]]` parameter to the task tool schema - The `start-batch`, `complete-batch`, and `reset-batch` actions require a list of task IDs - Previously, only `task_id` (singular string) was exposed in the MCP schema - Now both `task_id` and `task_ids` are available for single vs batch operations ## [0.8.16] - 2026-01-06 ### Changed - **Timeout standardization for AI CLI providers**: Increased default timeouts across all workflows - Per-provider/per-operation minimum: 360s (6 minutes) for Claude, Codex, Gemini, CursorAgent, OpenCode - Whole workflow timeouts: 600s (10 minutes) for plan_review, markdown_plan_review, deep_research - Deep research phase timeouts increased: planning/analysis/refinement 360s, synthesis 600s - Updated defaults in: `config.py`, `research.py`, `consensus.py`, `plan.py`, `review_helpers.py` - Updated sample and default TOML configs with new timeout standards and documentation ### Fixed - **Deep research `continue` action background execution**: Now properly supports `background=True` parameter - Previously, continuing a research session ignored the `background` flag - Added `background` and `task_timeout` parameters to `_continue_research()` method - Continued research can now run in background thread like initial `start` action - **Deep research status check crash**: Fixed `'NoneType' object has no attribute 'done'` error - `BackgroundTask.is_done` property now correctly handles thread-based execution - Previously used `task.done()` which only works for asyncio tasks - Now checks `thread.is_alive()` for thread-based execution (daemon threads with `asyncio.run()`) - Added comprehensive tests for background task state checking ## [0.8.15] - 2026-01-05 ### Fixed - **OpenCodeProvider server health check**: Now verifies server is actually responding, not just port open - Added `_is_opencode_server_healthy()` to verify HTTP response from `/session` endpoint - Prevents using stale processes or other services on port 4096 - Clear error message when port is in use by another process - **Missing `_intake_feature_flag_blocked` function**: Added missing function to authoring.py - Caused `NameError` when using intake-add, intake-list, intake-dismiss actions ## [0.8.14] - 2026-01-05 ### Fixed - **OpenCodeProvider silent server failure**: Empty responses no longer masked as SUCCESS - `opencode_wrapper.js` now exits with error (code 1) when server returns empty response - `opencode.py` validates non-empty content before returning SUCCESS status - Added logging to `_ensure_server_running()` for better visibility into server startup ## [0.8.13] - 2026-01-05 ### Fixed - **All CLI providers use stdin for prompts**: ClaudeProvider, GeminiProvider, and CursorAgentProvider now pass prompts via stdin - Avoids CLI argument length limits for long prompts - Updated `RunnerProtocol`, `_build_command()`, and `_run()` signatures to support `input_data` - Consistent pattern across all CLI-based providers (Claude, Gemini, Cursor, Codex) - **Full spec review collects all task files**: `_build_implementation_artifacts()` now collects `file_path` from all task/subtask/verify nodes - Previously only worked when task_id or phase_id was specified - Full spec reviews now include implementation artifacts from entire spec hierarchy ### Changed - **Provider tests updated**: Test assertions updated to reflect stdin-based prompt handling - `_build_command()` no longer includes prompt in returned command - Tests verify prompt is passed separately for stdin input ## [0.8.12] - 2026-01-04 ### Fixed - **CodexProvider stdin prompt handling**: Prompts are now passed via stdin instead of CLI arguments - Avoids CLI argument length limits for long prompts - Uses `-` marker to read prompt from stdin - Updated `_build_command()` and `_run()` signatures to support `input_data` ## [0.8.11] - 2026-01-04 ### Fixed - **Deep research background task execution**: Sub-queries now execute correctly - Root cause: `asyncio.create_task()` was called from sync MCP handlers without a running event loop - Solution: Use daemon threads with `asyncio.run()` for reliable background execution - `BackgroundTask` class now supports both thread-based and asyncio-based execution - **Intake-add priority validation UX**: Improved error messages and usability - Handle explicit `null` from JSON as "use default p2" (previously caused confusing error) - Error messages now include valid values: `p0, p1, p2, p3, p4` - Added human-readable priority aliases: `critical`, `high`, `medium`, `low`, `lowest` ## [0.8.10] - 2026-01-04 ### Added - **Implement command configuration**: New `[implement]` section in TOML config - `auto`: Skip prompts between tasks (default: false) - `delegate`: Use subagents for implementation (default: **true**) - `parallel`: Run subagents concurrently (default: false) - `model`: Size for delegated tasks - small, medium, large (default: small) ## [0.8.9] - 2026-01-04 ### Fixed - **Improved `phase-remove` discoverability**: Error message when attempting to remove a phase via `task action="remove"` now hints to use `authoring action="phase-remove"` instead ## [0.8.8] - 2026-01-04 ### Added - **ErrorCode.OPERATION_FAILED**: Added missing error code to `ErrorCode` enum - Fixes AttributeError when batch operations fail (e.g., `prepare-batch` on completed spec) ### Fixed - **Graceful `prepare-batch` on Complete Specs**: No longer throws error when spec is complete - Returns `{tasks: [], spec_complete: true}` instead of erroring - Enables clean detection of spec completion in parallel workflows ### Changed - **Test Updates**: Updated batch operation tests to verify graceful completion handling - `test_no_active_phases` now expects `None` error (not error string) - `test_prepare_batch_context_detects_spec_complete` verifies `spec_complete: true` response ## [0.8.7] - 2026-01-03 ### Added - **Research Nodes**: New `research` task type for spec-integrated research workflows - Add research nodes via `task(action="add", task_type="research", research_type="...", blocking_mode="...")` - Supported `research_type`: chat, consensus, thinkdeep, ideate, deep - `blocking_mode` controls dependency behavior: `none`, `soft` (default), `hard` - Research nodes with soft/none blocking don't block dependent tasks - **Spec-Integrated Research Actions**: New research tool actions for spec nodes - `node-execute`: Execute research workflow linked to spec node - `node-record`: Record research findings to spec node - `node-status`: Get research node status and linked session info - `node-findings`: Retrieve recorded findings from spec node - **Git Commit Suggestions**: Task completion now suggests commits based on git cadence - Response includes `suggest_commit`, `commit_scope`, `commit_message_hint` - Respects `[git].commit_cadence` config (task, phase, spec) - **Environment get-config Action**: Read configuration sections from foundry-mcp.toml - `environment(action="get-config", sections=["implement", "git"])` - Supports filtering by specific key within a section - **Research Memory Universal Lookup**: Added `load_session_by_id()` for loading any session type by ID prefix ### Fixed - **Spec Validation**: Added `failed` to allowed task statuses in `_validate_spec_structure()` - Batch operations set `status: "failed"` on task failure, but validation rejected it - Now allows: `pending`, `in_progress`, `completed`, `blocked`, `failed` - **Test Fixtures**: Fixed inline spec fixtures missing required `status` field on `spec-root` nodes ### Removed - **Obsolete Test**: Removed `test_flags_with_context` referencing removed feature flags system ## [0.8.6] - 2026-01-03 ### Fixed - **Spec Validation**: Added `failed` to allowed task statuses in `_validate_spec_structure()` - Batch operations set `status: "failed"` on task failure, but validation rejected it - Now allows: `pending`, `in_progress`, `completed`, `blocked`, `failed` - **Test Fixtures**: Fixed inline spec fixtures missing required `status` field on `spec-root` nodes - Updated 8 test specs in `test_batch_operations.py` to include `status: "in_progress"` ### Removed - **Obsolete Test**: Removed `test_flags_with_context` from `test_sdd_cli_runtime.py` - Test referenced `get_cli_flags()` and `flags_for_discovery()` from removed feature flags system ## [0.8.5] - 2026-01-03 ### Changed - **Default Disabled Tools**: Added `environment` to default `disabled_tools` list - Tools disabled by default: `error`, `metrics`, `health`, `environment` - These tools are only used during setup or for dashboard features ## [0.8.4] - 2026-01-03 ### Changed - **Tools Configuration Section**: Added dedicated `[tools]` config section for tool settings - `disabled_tools` now preferred under `[tools]` (backward compatible with `[server]`) - Comprehensive documentation of all available tools and their purposes - Updated sample config and setup template with `[tools]` section - Default recommendation: disable `error`, `metrics`, `health` to save context tokens ## [0.8.3] - 2026-01-03 ### Removed - **Feature Flags System**: Removed the entire feature flags subsystem to simplify codebase - Deleted `src/foundry_mcp/core/feature_flags.py` and `src/foundry_mcp/cli/flags.py` - Removed `dev_docs/mcp_best_practices/14-feature-flags.md` documentation - Server tools now always use unified manifest (previously feature-flag controlled) - Removed all feature flag imports and usage across tools ### Added - **Task Add `phase_id` Alias**: Added `phase_id` as alias for `parent` parameter in task `add` action - Provides more intuitive parameter name when adding tasks to phases - Falls back to `parent` if `phase_id` not provided ## [0.8.2] - 2026-01-03 ### Fixed - **CLI Provider Error Extraction**: Improved error message extraction from CLI tool outputs - `ClaudeProvider`: Extract errors from JSON output with `is_error: true` field - `CodexProvider`: Extract errors from JSONL events (`type: error`, `type: turn.failed`) - `CursorAgentProvider`: Check stdout for plain text errors (not just stderr) - `GeminiProvider`: Parse text + JSON error format, skip unhelpful `[object Object]` messages - `OpenCodeProvider`: Extract errors from wrapper JSONL output - **OpenCode NODE_PATH Configuration**: Added `_ensure_node_path()` to include global npm modules - Allows `@opencode-ai/sdk` to be installed globally rather than bundled - Detects global npm root via `npm root -g` and adds to NODE_PATH ## [0.8.1] - 2026-01-03 ### Added - **Batch Operations for Parallel Task Execution**: New actions for autonomous multi-task workflows - `prepare-batch`: Find independent tasks for parallel execution with file-path conflict detection - `start-batch`: Atomically start multiple tasks as in_progress (all-or-nothing validation) - `complete-batch`: Complete multiple tasks with partial failure support - `reset-batch`: Reset batch on failure, returning tasks to pending status - Token budget support for context-aware batch sizing - Stale task detection for tasks stuck in_progress beyond threshold - Dependency graph visualization in prepare-batch responses - **Autonomous Session Context Tracking**: CLI context management for continuous task processing - `AutonomousSession` class for tracking batch state across operations - `ContextTracker` for managing session lifecycle and context limits - Integration with task router for session-aware operations - **Configurable Tool Disabling**: Added `disabled_tools` configuration option to selectively disable MCP tools - Configure via environment variable: `FOUNDRY_MCP_DISABLED_TOOLS=error,health,metrics,test` - Configure via TOML: `[server] disabled_tools = ["error", "health", "metrics", "test"]` - Tools remain in codebase but are not registered with the MCP server when disabled - Useful for reducing context window usage by hiding unused tools ### Changed - Enhanced development guide with batch operations documentation and usage examples ## [0.7.11] - 2025-12-30 ### Fixed - **Consensus Workflow Provider Spec Parsing**: Extended provider spec parsing fix to consensus workflow - Full specs like `[cli]codex:gpt-5.2` in `consensus_providers` config now work correctly - Parses specs in `execute()`, `_query_provider_sync()`, and `_query_single_provider()` - Filters providers by base ID availability while preserving model selection ## [0.7.10] - 2025-12-30 ### Fixed - **Research Workflow Provider Spec Parsing**: Fixed `_resolve_provider` to handle full provider specs - Full specs like `[cli]codex:gpt-5.2-codex` in `default_provider` config now work correctly - Extracts base provider ID for availability check (e.g., `codex` from full spec) - Passes model from spec to `resolve_provider()` for proper model selection - Caches providers by full spec string to differentiate model variants ## [0.7.9] - 2025-12-30 ### Fixed - **Provider Detector Cache Isolation**: Fixed test pollution where availability cache persisted across tests - `reset_detectors()` now clears `_AVAILABILITY_CACHE` to ensure fresh detection - Prevents false negatives when test order affects cached availability results - **Research E2E Test Fixtures**: Added missing `max_messages_per_thread` to mock_config fixture - Fixed `TypeError: '>=' not supported between instances of 'MagicMock' and 'int'` - Research chat workflow now works correctly in test mode - **OpenCode Model Validation Test**: Removed obsolete test for empty model validation - Model validation was delegated to CLI in v0.7.5 but test was not removed - OpenCode provider correctly passes any model to CLI for validation ## [0.7.8] - 2025-12-30 ### Fixed - **Consensus Event Loop Conflict**: Fixed `asyncio.run() cannot be called from a running event loop` error - Replaced `asyncio.run()` with `ThreadPoolExecutor` for parallel provider execution - Works correctly within MCP server's event loop context - New `_execute_parallel_sync()` and `_query_provider_sync()` methods for thread-based parallelism - **Research Timeout Configuration**: Fixed thinkdeep and other workflows timing out after 30 seconds - Added `default_timeout` config option to `[research]` section (default: 60 seconds) - Workflows now use configurable timeout from config instead of hardcoded 30s - Longer-running investigation workflows like thinkdeep no longer timeout prematurely ## [0.7.7] - 2025-12-30 ### Added - **Research ProviderSpec Alignment**: Research config now supports full ProviderSpec notation like consultation - `default_provider` accepts both simple IDs (`"gemini"`) and ProviderSpec (`"[cli]gemini:gemini-2.5-flash"`) - `consensus_providers` accepts mixed notation for flexible model selection per provider - New `ResearchConfig.get_default_provider_spec()` helper parses default provider - New `ResearchConfig.get_consensus_provider_specs()` helper parses consensus providers - New `ProviderSpec.parse_flexible()` method for backward-compatible parsing - Workflows (`chat`, `consensus`, `thinkdeep`, `ideate`) now extract models from specs - Added `[research]` section to sample config with notation examples ## [0.7.6] - 2025-12-30 ### Fixed - **Research Tools Feature Flag**: Fixed bug where `research_tools = true` in `[features]` config section was ignored - Root cause 1: `research_tools` flag was never registered in the feature flag registry - Root cause 2: `[features]` section in TOML config was not being read - Added flag registration in `research.py` following `provider.py` pattern - Added global override support to `FeatureFlagRegistry` for config-based flag settings - Added `[features]` section handling in `ServerConfig._load_toml()` - Added `FOUNDRY_MCP_FEATURES` environment variable support (format: `flag1=true,flag2=false`) ### Added - **Feature Flag Global Overrides**: New methods on `FeatureFlagRegistry`: - `set_global_override(flag_name, enabled)` - Set config-based override for all clients - `clear_global_override(flag_name)` - Clear a global override - `clear_all_global_overrides()` - Clear all global overrides - `apply_config_overrides(features)` - Apply multiple overrides from config dict ### Dependencies - Added `filelock>=3.20.1` as a required dependency ## [Unreleased] ## [0.8.1] - 2026-01-03 ### Added - **Batch Operations for Parallel Task Execution**: New actions for autonomous multi-task workflows - `prepare-batch`: Find independent tasks for parallel execution with file-path conflict detection - `start-batch`: Atomically start multiple tasks as in_progress (all-or-nothing validation) - `complete-batch`: Complete multiple tasks with partial failure support - `reset-batch`: Reset batch on failure, returning tasks to pending status - Token budget support for context-aware batch sizing - Stale task detection for tasks stuck in_progress beyond threshold - Dependency graph visualization in prepare-batch responses - **Autonomous Session Context Tracking**: CLI context management for continuous task processing - `AutonomousSession` class for tracking batch state across operations - `ContextTracker` for managing session lifecycle and context limits - Integration with task router for session-aware operations ### Changed - Enhanced development guide with batch operations documentation and usage examples ## [0.8.0] - 2026-01-01 ### Added - **Deep Research Workflow**: Multi-phase iterative research with parallel source gathering - Background async execution with immediate `research_id` return for non-blocking operation - Five-phase pipeline: decomposition → search → analyze → synthesize → report - Query decomposition with strategic sub-query generation based on research intent - Gap detection and iterative refinement loops for comprehensive coverage - Final synthesis with confidence scoring (low/medium/high/confirmed) and source citations - Crash handler infrastructure for session recovery - Research tool actions: - `deep-research`: Start, continue, or resume research sessions - `deep-research-status`: Poll running research status with phase progress - `deep-research-report`: Get final markdown report with citations and audit trail - `deep-research-list`: List sessions with cursor-based pagination - `deep-research-delete`: Clean up research sessions - **Search Provider System**: Extensible multi-provider search architecture - **Google Custom Search**: Web search via Google's Custom Search JSON API - **Perplexity AI**: Sonar model integration for AI-powered search - **Tavily AI**: Search and extract modes with content analysis - **Semantic Scholar**: Academic paper search with citation metadata - Domain quality tiers (authoritative → unreliable) for source credibility scoring - Rate limiting and error handling per provider - Pluggable provider interface for custom backends - **Research Configuration**: New `[research.deep]` config section - Per-provider API key configuration (Google, Perplexity, Tavily, Semantic Scholar) - Timeout, iteration, and concurrency controls - Source quality and domain tier preferences - Storage path and session TTL settings - **Deep Research Documentation**: Comprehensive workflow documentation - `docs/concepts/deep_research_workflow.md`: Architecture and usage guide - `docs/examples/deep-research/`: Example research sessions and reports - **Provider Availability Caching**: Cache provider detection results to speed up MCP tool calls - New `[providers] availability_cache_ttl` config option (default: 3600 seconds) - Reduces repeated calls from ~5s to ~0s ### Changed - **Provider Model Validation Removed**: Model allowlists removed from all CLI providers - Providers no longer pre-register or validate model IDs against hardcoded lists - Any model string is now passed through to the underlying CLI for validation - Eliminates sync issues when providers release new models - Affected providers: `claude`, `gemini`, `codex`, `cursor-agent` - Default models remain as fallbacks: opus, pro, gpt-5.2, composer-1 - **BREAKING: Simplified Spec Templates**: Removed pre-baked spec templates (simple, medium, complex, security) - Only `empty` template is now supported - creates a blank spec with no phases - Use phase templates (`planning`, `implementation`, `testing`, `security`, `documentation`) to add structure - Default template changed from `medium` to `empty` - Mission statement no longer required (was required for medium/complex) - `_requires_rich_task_fields()` now checks explicit `complexity` metadata instead of template - Passing deprecated templates (simple, medium, complex, security) returns validation error ### Fixed - **AI Consultation Config Loading**: Fixed issue where AI consultation features returned `model_used: "none"` because config was loaded from CWD instead of workspace path - `review.py`: Now loads `foundry-mcp.toml` from workspace path for fidelity reviews - `plan.py`: Added `_find_config_file()` helper to walk up directories and find config ### Migration **Spec Templates:** ```python # Old approach (no longer works) authoring(action="spec-create", name="my-feature", template="medium", mission="...") # New approach authoring(action="spec-create", name="my-feature") authoring(action="phase-template", template_action="apply", template_name="planning", spec_id="...") authoring(action="phase-template", template_action="apply", template_name="implementation", spec_id="...") ``` **Deep Research:** ```python # Start a deep research session (returns immediately with research_id) research(action="deep-research", query="What are the best practices for LLM evaluation?") # Poll for status research(action="deep-research-status", research_id="...") # Get final report when complete research(action="deep-research-report", research_id="...") ``` ## [0.7.0] - 2025-12-30 ### Added - **Research Router**: New unified research tool providing multi-model orchestration capabilities - **chat**: Single-model conversation with thread persistence - Thread creation with title and system prompt - Conversation continuation via thread_id - Token budgeting for context management - Thread CRUD operations (list, get, delete) - **consensus**: Multi-model parallel consultation with synthesis - Parallel execution via asyncio.gather with semaphore limiting - Four synthesis strategies: all_responses, synthesize, majority, first_valid - Partial failure handling with min_responses and require_all options - Configurable timeout per provider - **thinkdeep**: Hypothesis-driven systematic investigation - Investigation step execution with state persistence - Hypothesis creation and tracking with evidence accumulation - Confidence level progression (speculation -> confirmed) - Convergence detection based on depth and confidence - **ideate**: Creative brainstorming with idea clustering - Four-phase workflow: divergent, convergent, selection, elaboration - Multi-perspective idea generation - Automatic clustering and scoring - Detailed plan elaboration for selected clusters - **ResearchConfig**: New configuration section for research workflows - Configurable storage path, TTL, max messages per thread - Default provider and consensus provider list - ThinkDeep max depth and Ideate perspectives - **Research Data Models**: Pydantic models for all workflow states - Enums: WorkflowType, ConfidenceLevel, ConsensusStrategy, ThreadStatus, IdeationPhase - Conversation, ThinkDeep, Ideate, and Consensus state models - **File-Based Memory Storage**: Persistent state management for research sessions - FileStorageBackend with CRUD operations - File locking via filelock for thread safety - TTL-based cleanup for expired sessions - **Research Test Suite**: 149 tests covering models, memory, and router - **Feature Flag**: `research_tools` flag (experimental) gates research tool access ## [0.6.0] - 2025-12-29 ### Added - **Notes Intake System**: Fast-capture queue for rapid idea/task capture with automatic triage workflow - **intake-add**: Add items to the intake queue with title, description, priority (p0-p4), tags, source, and requester fields - Idempotency key support for deduplication (checks last 100 items) - Tag normalization to lowercase - Full dry-run support for validation without persistence - **intake-list**: List pending intake items in FIFO order with cursor-based pagination - Configurable page size (1-200, default 50) - Efficient line-hint seeking with fallback to full scan - Returns total_count for queue size visibility - **intake-dismiss**: Mark items as dismissed with optional reason - Atomic file rewrite pattern for data integrity - Supports dry-run mode - JSONL-based storage at `specs/.notes/intake.jsonl` with fcntl file locking - Automatic file rotation at 1000 items or 1MB - Thread-safe and cross-process safe with 5-second lock timeout - Security hardening: path traversal prevention, prompt injection sanitization, control character stripping - Feature flag gated: `intake_tools` (experimental, opt-in) - **Intake Schema**: JSON Schema for intake-v1 format with comprehensive validation constraints - **Intake Documentation**: User guide at `docs/guides/intake.md` - **RESOURCE_BUSY Error Code**: New error code for lock contention scenarios ## [0.5.1] - 2025-12-27 ### Added - **Phase Metadata Updates**: New `authoring action=phase-update-metadata` for updating phase-level metadata - Supports updating `estimated_hours`, `description`, and `purpose` fields - Full dry-run support for previewing changes - Tracks previous values for audit purposes - Core function `update_phase_metadata()` in `spec.py` with comprehensive validation ### Fixed - **Lifecycle Tool Router Compatibility**: Fixed `_handle_move()` and other lifecycle handlers receiving unexpected keyword arguments (`force`, `to_folder`) from the unified router dispatch - All lifecycle handlers now accept full parameter set for router compatibility - Resolves errors like `_handle_move() got an unexpected keyword argument 'force'` ## [0.5.0] - 2025-12-27 ### Added - **Spec Modification Capabilities**: Complete implementation of dynamic spec modification (7 phases, 54 tasks) - **Task Hierarchy Mutations**: `task action=move` for repositioning tasks within/across phases with circular reference prevention - **Dependency Management**: `task action=add-dependency`, `task action=remove-dependency` for blocks/blocked_by/depends relationships - **Task Requirements**: `task action=add-requirement` for adding structured requirements to tasks - **Bulk Operations**: `authoring action=phase-add-bulk` for batch phase creation, `authoring action=phase-template` for applying predefined structures - **Metadata Batch Updates**: `task action=metadata-batch` with AND-based filtering by node_type, phase_id, or pattern regex - **Find-Replace**: `authoring action=spec-find-replace` with regex support and scope filtering for bulk spec modifications - **Spec Rollback**: `authoring action=spec-rollback` for restoring specs from automatic backups - **Spec History & Diff**: `spec action=history` for backup timeline, `spec action=diff` for comparing specs - **Validation Enhancements**: `spec action=completeness-check` with weighted scoring (0-100), `spec action=duplicate-detection` with configurable similarity threshold - **Standardized Error Codes**: New `ErrorCode` enum with semantic error codes per 07-error-semantics.md - **Contract Tests**: Comprehensive test suite for response-v2 envelope compliance across all phases ### Changed - Updated capabilities manifest with 15 new actions documented - Spec modification spec moved from pending to active (100% complete) ## [0.4.2] - 2025-12-24 ### Added - **Preflight Validation**: `authoring action=spec-create dry_run=true` now generates and validates the full spec, returning `is_valid`, `error_count`, `warning_count`, and detailed diagnostics before actual creation - **Schema Introspection**: New `spec action=schema` returns all valid enum values (templates, node_types, statuses, task_categories, verification_types, journal_entry_types, blocker_types, status_folders) for LLM/client discovery ### Changed - **Spec Field Requirements**: Medium/complex specs now require `metadata.mission` and task metadata for `task_category`, `description`, and `acceptance_criteria`; implementation/refactoring tasks must include `file_path` - **Task Metadata Updates**: `task update-metadata` now accepts `acceptance_criteria` and aligns task category validation with the canonical spec categories ## [0.4.1] - 2025-12-24 ### Added - **Batch Metadata Utilities**: New task actions for bulk operations - `task action=metadata-batch`: Apply metadata updates to multiple nodes with AND-based filtering by `node_type`, `phase_id`, or `pattern` regex - `task action=fix-verification-types`: Auto-fix invalid/missing verification types on verify nodes with legacy mapping support - Both actions support `dry_run` mode for previewing changes - **Phase-First Authoring**: New `authoring action=phase-add-bulk` for creating multiple phases at once with metadata defaults - **Spec Mission Field**: Added `mission` field to spec metadata schema for concise goal statements - **Workflow Timeout Override**: AI consultation now supports workflow-specific timeout configuration ### Changed - **JSON Output Optimization**: CLI and MCP server now emit minified JSON (no indentation) for smaller payloads - **Fidelity Review Improvements**: Better path resolution with workspace_root support, graceful handling of non-JSON provider responses - **Provider Configuration**: Updated OpenCode model IDs and default model; reordered provider priority - **Claude Provider Tests**: Updated to use Haiku model for faster test execution ### Fixed - Fixed parameter filtering in error_list handler to prevent unexpected argument errors - Fixed duplicate file paths in fidelity review implementation artifacts - Synced `__init__.py` version with `pyproject.toml` ## [0.4.0] - 2025-12-23 ### Changed - **Verification Types**: Aligned task API and spec validator to use canonical values (`run-tests`, `fidelity`, `manual`) - Task API now accepts `run-tests`, `fidelity`, `manual` (previously `auto`, `manual`, `none`) - Spec validator updated to match canonical schema values - Legacy values automatically mapped: `test` → `run-tests`, `auto` → `run-tests` ### Added - **Auto-fix for `INVALID_VERIFICATION_TYPE`**: Specs with legacy verification types are now auto-fixable via `validate-fix` - **Auto-fix for `INVALID_ROOT_PARENT`**: Specs where spec-root has non-null parent are now auto-fixable ### Removed - Removed `foundry-mcp-ctl` package and mode-toggling feature - server now always runs with all tools registered ## [0.3.4] - 2025-12-21 _Note: Mode toggling features added in this version were subsequently removed._ ## [0.3.3] - 2025-12-17 ### Changed - **Dashboard**: Refactored pages module to views with cleaner organization - **Dashboard**: Improved data stores with better caching and filtering - **Observability**: Added action label to tool metrics for router-level granularity - **Providers**: Codex CLI now ignores unsupported parameters (warning instead of error) ### Added - Dashboard PID file tracking for cross-CLI process management - Tool usage dashboard view with action-level breakdown - OpenCode Node.js wrapper for subprocess execution - Integration tests for provider smoke testing, fidelity review flow, and plan review flow ### Fixed - Codex provider environment handling (unsets OPENAI_API_KEY/OPENAI_BASE_URL that interfere with CLI) - Minor fixes to Claude and Gemini providers ## [0.3.2] - 2025-12-16 ### Added - Launcher script (`bin/foundry-mcp`) for configurable Python interpreter selection - `FOUNDRY_MCP_PYTHON` environment variable to override the default Python interpreter ### Fixed - Removed duplicate `spec_id` and `node_id` fields from task progress response ## [0.3.1] - 2025-12-16 ### Removed - Removed `code` unified tool (find-class, find-function, callers, callees, trace, impact actions) from MCP surface. Unified manifest reduced from 17 to 16 tools. ## [0.3.0] - 2025-12-15 ### Changed - Consolidated the MCP tool surface into 17 unified routers (tool + `action`) and aligned CLI/MCP naming. - Updated documentation and manifests to reflect the unified router contract. ### Added - New completed specs documenting MCP tool consolidation and removal of docquery/rendering/docgen. - Unified-manifest budget telemetry (Prometheus metrics, recording rules, alerting rules, and dashboard panels). ### Removed - Legacy per-tool MCP modules and legacy CLI command surfaces in favor of unified routers. - Docquery/rendering/docgen modules and generated docs previously under `docs/generated/`. ## [0.2.1] - 2025-12-08 ### Changed - **Dashboard**: Replaced aiohttp+vanilla JS dashboard with Streamlit for better visualizations and interactivity - Dashboard dependencies changed from `aiohttp` to `streamlit`, `plotly`, `pandas` - Default dashboard port changed from 8080 to 8501 (Streamlit default) ### Added - New Streamlit dashboard with 5 pages: Overview, Errors, Metrics, Providers, SDD Workflow - Interactive Plotly charts with zoom, pan, and hover tooltips - Data export functionality (CSV/JSON download buttons) - Cached data access via `@st.cache_data` for performance - CLI commands: `dashboard start`, `dashboard stop`, `dashboard status` - New SDD Workflow page for spec progress tracking, phase burndown, task status - Plan review tool (`plan-review`) for AI-assisted specification review ### Removed - Old aiohttp-based dashboard server and static JS/CSS files