Changelog

This file tracks repository-level release history for code2skill. Detailed notes for each tagged release live under docs/releases/.

Unreleased

Added a reproducible structural evidence benchmark with path-only and AST-symbol baselines, README chart, result JSON, and benchmark notes.
Fixed Python entrypoint role scoring so main.py and similar entry files are not mislabeled as root configuration files.
Added semantic Python extraction for call targets, dynamic imports, type references, raised exceptions, class attributes, and lightweight data-flow edges.
Improved internal import resolution for detailed from ... import ... records and dynamic imports.
Fed call/type/data-flow evidence into file prioritization, planning prompts, generated skeletons, and project summaries.
Added algorithm notes documenting the paper-backed ideas behind the scanner improvements.
Added doctor readiness checks for generated bundles, Skill plans, state snapshots, and adapted target files.
Added repository-specific adoption-guide.md output and updated README/docs around first adoption, CI refresh, and multi-tool publishing workflows.
Changed merge-style adapters to preserve hand-written content through a managed code2skill block.
Added Cursor copy-target manifest cleanup so stale generated rules are removed while unmanaged team rules are preserved.
Added OpenAI-compatible Responses endpoint configuration through CODE2SKILL_OPENAI_BASE_URL.
Improved Skill planning and writing prompts to avoid documentation/packaging-only pseudo-skills and produce more maintainer-oriented guidance.
Improved PyPI-facing README content, package metadata, and sdist documentation inclusion.
Added explicit user personas and business scenarios to README/use-case documentation.
Clarified English/Chinese README entry points and rewrote public docs with more direct maintainer-facing language.
Hardened adapt to reject incomplete generated Skill directories before writing target files.
Hardened doctor --target cursor to require the copy manifest used for stale generated-rule cleanup.
Hardened CLI environment variable parsing to report invalid values instead of silently falling back to defaults.
Restricted sdist docs inclusion to public docs and release notes, excluding internal planning docs.
Fixed cost-estimate assumption text that rendered as mojibake in CLI/report output.
Fixed LLM invalid-JSON responses to produce a clear runtime error.
Fixed incremental generation to replan when affected Skills are missing from the existing plan.