Author, check, and critically review AAanalysis docstrings against the house style of the mature classes (CPP, dPULearn, AAclust). Bundles a deterministic structural checker (scripts/check_docstrings.py — numpydoc shape, named Returns, per-method Examples include, cross-reference integrity, citation-defined, stub-skipping) and a doc-vs-signature drift detector (scripts/doc_signature_drift.py — documented default/type vs the real signature). For deep reviews it fans out per-subpackage reviewers, cross-checks each docstring against the implementation + sibling docstrings + the CONTEXT.md glossary for clarity and consistency, classifies fixes as Necessary / Nice-to-have / Can-be, optionally refines the plan with grill-with-docs, then refactors staged or in one pass. Use when writing or reviewing the docstring of any public class/method/function under aaanalysis/, when a new class must match conventions, or when the user asks to audit, grade, lint, or critically review docstring / API-doc quality or find doc-vs-c
Autonomously drive an entire issue or a complete new feature from spec to merge through the AAanalysis agentic-engineering protocol — sharpen the issue, branch into an isolated worktree, implement, push and open the PR early so CI runs, run the automated review + quality gates, keep the branch synced, then hold at the human-review gate where the user picks a manual PR-review loop or skips (skip = post an approving review comment, then arm GitHub-native auto-merge), with a fix-forward loop on red CI. Use when the user wants to start work on an issue, "walk me through the workflow", take a change from issue to merge, or wants the auto-merge / auto-fix loop driven for them. NOT for typos, one-line fixes, or trivial local edits — make those directly. The canonical protocol lives in docs/guides/agentic_engineering.md; this skill executes it.
Audit all open GitHub issues for the aaanalysis package against its scope and coding standards (CLAUDE.md + .claude/rules/sharp-edges.md), flagging each as ready / needs-revision / conflicts / already-addressed, detecting overlaps that must not be developed in parallel, and writing a prioritized step-by-step implementation plan to docs/guides/handoff_github_issues.md. Issues are the primary lens; pull requests are mapped in as secondary context — an "Issue ↔ PR activity" table (issues on the left, their PR(s) on the right, — when an issue has no PR) plus per-issue in-flight-PR notes so each issue and its PR are shown together. Use when the user wants to review, triage, audit, or scope GitHub issues or PRs, refresh the issue handoff, see what work is in flight, decide what to implement next, or plan parallel work across issues. Also use when the user wants to create a new GitHub issue to the house style guide — newly created issues are written to docs/guides/issue_style_guide.md and then added to the Kanban Pr