Standardized DESCRIPTION file formatting to fix a total breakdown in the pkgstats function call network report.
Removed the manual Collate field to reduce metadata complexity and allow default alphabetical loading order (compatible with zzz.R initialization).
Replaced bitwise control characters (\001, \002) in mermaid_parser.R with safe string delimiters (@@@, ###) to prevent signal interruptions in static analysis parsers.
Workflow & Automation:
Updated the pkgdown GitHub workflow to ensure unit tests are executed as a prerequisite for documentation deployment.
Integrated the official ropensci/pkgcheck-action workflow to automate rOpenSci compliance checks on every push.
Standardized Roxygen2 annotations across all core R6 classes (AgentDAG, AgentNode, AgentDriver, MessageLog, Checkpointer, etc.).
Added detailed, type-annotated @param descriptions for all public methods.
Implemented robust, non-interactive-safe @examples wrapped in \dontrun{} to demonstrate complex multi-step workflows.
Added explicit environment variable setup instructions (.Renviron) for all API-driven components.
Workflow Lifecycle: Documented the full “Low Code” lifecycle from declarative Mermaid/YAML definitions to compiled and executed DAGs.
Improved Driver Registry: Improved discovery and hot-swapping documentation for LLM drivers.
Persistence & Auditing: Added lifecycle documentation for DuckDB and JSONL message logging and state checkpointing.
Bug Fixes & Stabilization:
Resolved a syntax error in GeminiImageDriver.
Fixed Roxygen parsing errors where method titles merged into descriptions in man/AgentDAG.Rd.
Restored flexible function lookup in auto_node_factory to maintain backward compatibility with global functions in tests.
Clean Registry: Package now achieves a perfectly clean R CMD check (0 Errors, 0 Warnings, 0 Notes) on core builds.
Complete Instruction Manual: Rewrote vignettes/manual.md as a 16-part narrative-style beginner guide, progressing from a single “hello world” node through loops, YAML workflows, checkpointing, and git worktree isolation. Includes a direct side-by-side comparison showing the YAML-first approach reduces a 40-line R script to 4 lines.
Test Hygiene: Silenced expected diagnostic warnings across the test suite for cyclic and multi-root graph patterns. Removed deprecated testthat::context() calls.
HydraR 0.2.1 (Final Polish & Acknowledgements)
Formal Acknowledgements: Added a dedicated section to README.md and paper.md acknowledging funding from Bioplatforms Australia via NCRIS and the Australian Proteome Analysis Facility (APAF).
Institutional Branding: Integrated high-quality logo assets for APAF, Macquarie University, Bioplatforms Australia, NCRIS, and NATA into the documentation.
Accreditation Detail: Added formal NATA accreditation phrase (ISO/IEC 17025) to Acknowledgements.
Improved CLI Driver Resilience:
Updated OpenAICodexCLIDriver to support --skip-git-repo-check, enabling reliable operation in non-Git temporary environments (e.g., R CMD check).
Fixed AnthropicCLIDriver test failures by automatically applying --dangerously-skip-permissions during verification.
R CMD Check Cleanliness: Resolved final test errors and silenced diagnostic warnings to achieve a 100% clean check state.
Markdown Vignettes Integration: Introduced dedicated markdown versions of core case studies (hong_kong_travel.md, sorting_benchmark.md, state_persistence.md, extending_hydrar.md, targets_integration.md) for easier discovery on GitHub and documentation sites.
HydraR 0.2.0.9000 (Standardized Drivers)
Standardized Driver Architecture: Renamed all core AI drivers for internal consistency and Provider + Mode clarity (e.g., AnthropicCLIDriver, AnthropicAPIDriver, OpenAIAPIDriver).
OpenAI Shorthand Refactor: Swapped driver=openai to favor the high-performance Codex CLI as the default, with driver=openai_api for cloud reasoning.
Anthropic Shorthand Update: Renamed driver=claude to driver=anthropic to match provider-first naming standards.
Improved GPT-5.4 Support: Updated the OpenAIAPIDriver defaults for seamless integration with GPT-5.4 flagship/mini models.
R6 Standards: Resolved internal finalize() warnings to ensure full compliance with R6 2.4.0+ and suppressed developmental noise.
HydraR 0.1.0 (Initial Release)
Agentic Orchestration: Implementation of the AgentDAG R6 class for managing complex directed workflows and iterative loops (state machines).
Hardened State Management: Introduced the AgentState system, transitioning from nested R lists to a robust key-value store with support for complex reducers and history tracking.
Provider-Agnostic Drivers: Initial support for GeminiCLIDriver, AnthropicCLIDriver, and CopilotCLIDriver, enabling “hot-swappable” LLM backends.
Git Worktree Isolation: Developed the worktree.R module to provide safe, parallel execution environments for file-modifying agentic workflows.
High-Fidelity Validation: Integrated Auditor and Harmonizer patterns for autonomous quality control and merging of distributed agent outputs.
Persistence & Checkpointing: DuckDB and SQLite-backed AgentCheckpointer implemented for resumable and auditable execution.
Mermaid Round-Trip: Visual design engine for converting between R objects and Mermaid.js syntax, enabling human-in-the-loop auditability.
AI Transparency: Disclosure of LLM usage in development through agents.md and DESIGN.md.