Changelog

Source: NEWS.md

kagiPro 0.4.1

Documentation

• Externalized the generic r-package-developer skill from package content to repository-level skills/r-package-developer.
• Updated skill discovery guidance to clearly distinguish package-bundled skills (inst/skills) from the external generic developer skill.
• Added explicit policy text: if the external skill is missing, suggest installation, but only install with explicit developer approval and never automatically.
• Removed the pkgdown wrapper page for the external developer skill and updated pkgdown article/menu configuration accordingly.

kagiPro 0.4.0

Features

• Added kagi_fetch() as a high-level project-folder workflow helper with endpoint-scoped outputs (<project>/<endpoint>/json and <project>/<endpoint>/parquet).
• Added endpoint-specific query support for FastGPT via query_fastgpt().
• Added modular corpus enrichment pipeline for search-like corpora:
  • download_content()
  • content_markdown()
  • markdown_abstract() with endpoint/query partitioned output under <project>/<endpoint>/{content,markdown,abstract}.
• Added pluggable text summarization backends:
  • summarize_with_openai()
  • summarize_with_kagi() (text-based summarizer usage).
• Added read_corpus() with optional abstract linking (abstracts = TRUE) by id + query.
• Standardized query constructor outputs to named lists across endpoints.
• Extended kagi_request() with error_mode ("stop", "write_dummy") for strict vs resilient execution modes.
• Improved handling of mixed query lists so long batch runs can continue with structured dummy outputs on failure.
• Kept JSON-to-parquet as a first-class path via kagi_request_parquet() for downstream analytics.
• Added low-level 00_in.progress marker lifecycle to request/parquet runs for folder-state visibility during long jobs.
• Added metadata persistence in kagi_request() for replayable query runs:
  • per-query _query_meta.json
• Added kagi_update_query() to rerun by query_name across matching endpoints and refresh only touched parquet query=<name> partitions.
• Added clean_request() to remove JSON request artifacts project-wide while preserving _query_meta.json per query (with dry_run space estimates).
• Added AI-agent skill scaffolding under inst/skills:
  • maintainer-workflow
  • user-search, user-enrich, user-summarize, user-fastgpt

Bug Fixes

• Fixed inconsistent request handling between single-query and list-query execution paths.
• Fixed recursive/list request execution to propagate request options (including error handling mode) correctly.
• Fixed output path behavior so single-item list requests do not create unexpected nested output directories.
• Removed redundant API key resolution calls in connection flow where applicable.
• Fixed fixture-location inconsistencies in tests by using a single cassette location.

Breaking Changes

• Renamed query constructors to query_<endpoint> for discoverability:
  • search_query() -> query_search()
  • enrich_web_query() -> query_enrich_web()
  • enrich_news_query() -> query_enrich_news()
  • summarize_query() -> query_summarize()
  • fastgpt_query() -> query_fastgpt()
• Removed add_sbstract_to_parquet(). Abstract creation is now handled only via the modular content pipeline (download_content() -> content_markdown() -> markdown_abstract()).
• kagi_request_parquet() is JSON-to-parquet conversion only and no longer accepts abstract-augmentation arguments.
• Query constructors consistently return named lists. Code that previously assumed a bare single query object may require [[1]] indexing in some direct calls.
• Error behavior can now be configured explicitly; strict failure remains default, but resilient mode writes dummy payloads and warnings instead of stopping.

Documentation

• Updated package docs to align with current function names and design principles.
• Reworked kagi_request documentation to remove outdated references and clarify endpoint behavior.
• Expanded vignette set:
  • Quickstart guide
  • Search endpoint guide
  • Enrich endpoint guide
  • Summarize endpoint guide
  • FastGPT endpoint guide
• Shifted vignettes toward user-oriented narrative style while retaining runnable code examples.
• Updated pkgdown article structure for clearer endpoint-based navigation.
• Added project-level design/context documentation for maintainers and AI coding agents (PROJECT_DESIGN.md).
• Added README disclaimer and AI-assisted development notice.
• Added AI-readable artifact index files (llms.txt, llms-full.txt) and mirrored pkgdown extras.
• Added AI-focused vignettes:
  • agent-quick-index
  • api-contracts
• Added rendered Skills pages in pkgdown using include-based wrappers (vignettes/skills-*.qmd) with one page per skill and embedded references.

Maintenance

• Added CI workflow for package checks on pull requests to main and pushes to dev.
• Removed unused legacy package assets from inst/ (extdata, plantuml, query_test.R).
• Added AI-doc consistency check script (scripts/check-ai-docs.sh) and wired it into CI (R-CMD-check).
• Added generic reusable maintainer skill r-package-developer with:
  • branch protection baseline,
  • release/validation checklist,
  • commit template,
  • skill design standard.
• Standardized all skill files to a unified structure with a dedicated ## References section.