Changelog

All notable changes to this project are documented here. The format follows Keep a Changelog, and the project adheres to Semantic Versioning.

[Unreleased]

dp check command — reports whether the README or any docstrings are stale and exits non-zero if so, with no model calls or configuration required, so it can run as a pre-commit hook. dp init runs the same report at the end of setup.
Welcome banner — dp init opens with an ASCII banner showing the version and issue link; the same banner heads the project README.
dp readme command and README pipeline: a LangGraph agent scopes a directory, maps its source with a fast analysis model, then drafts or refreshes a README.md for side-by-side review before writing.
Shared ChangeManifest backing both pipelines — semantic-hash change detection with a stat-stamp fast-skip, keyed per pipeline.
DocstringGenerator protocol so the docs pipeline depends on a generator interface rather than the concrete LLMDocstringGenerator.
py.typed marker — the package now ships its type information to consumers.
Persistent run timer: a single elapsed counter ticks for the whole docs and readme run, stepping aside only for interactive prompts.
Status message while the docs review reruns docstrings, so the regeneration step is no longer silent.
PyPI publishing metadata — SPDX license, classifiers, keywords, and project URLs.

Shared confirm_or_skip helper backs the docs and readme pre-run prompts, so both gate model calls the same way.
Docs run summary formats elapsed time with format_duration, matching the readme summary and the live timer.

mypy moved from runtime dependencies to the dev group — it was never imported at runtime and shipping it added ~50 MB to every install.
Dead, test-only helpers: GitReader.is_repo / last_commit_touching / commits_since and the BatchPlan count properties.

ConfigError construction routed through named factory methods (conflicting_flags, no_runs_to_resume) instead of inline strings.

dp CLI with init, docs, cleanup, and config commands.
Docs pipeline: scout undocumented code, generate Google-style docstrings in batched LLM calls, and insert them without disturbing surrounding source.
Anthropic, OpenAI, and Gemini providers with fast / balanced / best model tiers.
Anti-LLM-ese retries, transient-failure retries, and on-exhaustion model switching.
Checkpointed runs with --resume, plus a per-run manifest of counts, tokens, and cost.
Fast-skip cache so already-documented, unchanged files are not re-sent.
Interactive review panel for accepting, rejecting, or rerunning generated docstrings.
Layered config (defaults < global < repo) with masked API keys and an error hierarchy carrying actionable hints.