argythana

Use this skill when the user wants an iterative `code-review` loop over Python changes: review a Python diff, PR, branch, or commit range with `code-review`, fix the actionable Critical, High, and Moderate findings, run the relevant checks and tests, and repeat until 5 consecutive clean PASS reviews, `insufficient evidence`, or max cycles. Trigger on "run code-review until clean", "keep reviewing and fixing this Python PR until it passes", or "fix the code-review findings and rerun review until acceptable". If the target repo, diff, branch, commit, or commit range is referenced but unavailable, ask after routing. Do NOT use for a single review-only pass; use `code-review`. Do NOT use for pytest-suite-specific audit loops; use `pytest-suite-review-loop`. Do NOT use for implementing a task against Acceptance Criteria; use `task-implementation-loop`. Do NOT use for non-Python code, dependency-only changes, or writing new features from scratch.

Use this skill when the user wants a production-grade review of Python code, a Python pull request, or a Python-focused diff. Check architecture, security, code quality, tests, documentation, deployment risk, and code-documentation consistency. Trigger on requests like "review this Python PR", "audit this diff", "check code quality", or "give me a fix plan". Do NOT use it for non-Python code review, writing new features, dependency-only changes, pytest-suite-specific audits where `pytest-suite-review` is a better fit, or iterative review-fix-rerun loops where `code-review-loop` is a better fit.

Use when the user wants to import a local Markdown file as a new note into a personal Obsidian vault — prompts like "import this .md into Obsidian", "add this Markdown doc to my vault", or "create a note from this file". The vault root is discovered at runtime: `OBSIDIAN_VAULT` env var if set, otherwise the single directory under `$HOME/Documents` (depth 3) containing `.obsidian/`. If the source Markdown path is missing, ask for it. Do NOT use for: editing, renaming, moving, or deleting existing notes; vault-wide search; daily notes (`daily/`); Literature notes (`Literature/`); ephemeral scratch (`ephemeral/`); Excalidraw drawings; fetching from a URL (use `url-reader` first); or authoring note content from scratch. Do NOT run `git` — the `obsidian-git` plugin owns commits. Assumes the layout in `references/vault-conventions.md` (`all_notes/`, `tags/`, etc.).

Use when expanding, annotating, rewriting, or drafting a section or whole notebook of an official tutorial for a Python data/ML library (e.g. scikit-learn, PyTorch, pandas, matplotlib, NumPy, SciPy, Plotly) so it is friendlier for beginner data scientists and CS students. Trigger on prompts like "expand this scikit-learn section", "explain this pandas concept better", "make this PyTorch notebook beginner-friendly", or "why does this feature exist". If the target notebook or cell is unclear, ask after routing. Do NOT use for general Python tutorial writing unrelated to data/ML libraries (Django, Flask, application code), editing the upstream library itself, non-tutorial docs (`README.md`, `CLAUDE.md`), or non-teaching notebook changes (file moves, dependency bumps, formatting, kernel metadata). If a library-specific tutorial-expand skill is installed (e.g. `mlflow-tutorial-expand`), prefer that. For deps use `uv-deps`; for staging use `git-staging`; for messages use `commit-message`.

Use this skill when the user wants to lazily import Python modules, defer module loading until first use, or reduce Python or CLI startup cost by changing import behavior. Trigger on requests like "import X lazily", "don't load pandas until it's used", "use PEP 810", "use `wrapt.lazy_import()`", or "make this CLI's imports lazy". If the target code or supported Python version is missing and cannot be inferred from the repo, ask for it after routing. Do NOT use it just because a Python file has heavy dependencies, or for general import cleanup, generic startup tuning that is not about imports, or dependency-management requests where the main task is adding or updating packages; use `uv-deps` for that work.

Use when the task is to manage dependencies in a `uv` project — either a single-package project (one `pyproject.toml`) or a `uv` workspace monorepo (multiple per-package `pyproject.toml` files plus a root `uv.lock`). Covers add, remove, update, sync, lock, build, export, or reorganize packages, dependency groups, extras, and workspace sources via `uv` commands rather than hand-editing `pyproject.toml`. Trigger for requests like 'add a package', 'uv sync', 'move this to dev dependencies', 'update uv deps', 'remove X from package Y', 'regenerate constraints.txt', or 'build a wheel'. Do NOT use for Poetry/pip/conda/pip-tools workflows, manual virtualenv setup or activation, import sorting, or unrelated `pyproject.toml` edits such as project metadata or build config.

Use this skill when the user asks for a verifiable literature review or peer-reviewed citations for an ADR, design doc, paper, or technical decision, with prompts like "literature review on X", "find papers on X", "survey the literature on X", or "cite peer-reviewed sources for X". Enforce structural verification: every emitted citation needs metadata lookup, full-text retrieval, a supporting quote with locator, byte-for-byte author attribution, Negative Findings, and a self-audit table. Do NOT use it for single-URL summarization, drafting the surrounding ADR, casual citation pointers where verification is explicitly waived, fictional references, or unsourced factual questions.

Use this skill when the user wants to inspect one local tabular file (Parquet, CSV, JSON, JSONL) — verify the file is readable, list columns and types, count rows, report file size, or produce a compact schema/connection report. Trigger on prompts like "inspect this dataset", "check the schema", "how many rows are in this file", "verify data access", or "what columns are in this parquet". If the file path is missing, ask for it after routing. Do NOT use it for exploratory statistics, distributions, missing patterns, or class balance; use `eda` after access is confirmed. Do NOT use it for model training, database connections (Postgres, MySQL, SQLite), remote data sources (S3, HTTPS, Snowflake, BigQuery), or non-tabular data (images, text, audio).

Use this skill to give a student FORMATIVE, instructor-voice feedback on a draft (often partial) of the Python-course final assignment — guidance on how to improve, never a grade. Trigger on requests like "give feedback on this assignment draft", "review <student>'s final assignment submission", "feedback on the eClass submission for <student>", "the student submitted an early draft, what should they improve", "feedback on this regression/clustering/classification notebook for the final assignment", or when the user supplies a .zip / folder / .ipynb of a final-assignment draft and asks for feedback rather than a grade. Both the eClass download automation (automation_infrastructure/eclass/download_submissions.py) and emailed drafts land in one unified per-student location, students_work/class_<YY>/<lastname_t>/final_assignment/; the skill locates the student there, reads the already-extracted dated download (or extracts a .zip only if needed). It reads the authoritative specs (final_assignment/submission_requi

Use this skill to draft English-language LinkedIn posts announcing updates to the UoA "Python for Data Science, ML and AI" MSc course. Trigger on requests like "draft the next linkedin post", "write the linkedin post about <theme>", "linkedin post for the course updates", "continue the course post series", "mark post N as published", or "rebuild the linkedin series plan". The series plan (themes, evidence, status) is the file admin_docs/linkedin_posts/series_plan.md — this skill reads it, gathers and VERIFIES evidence from git history and repo files, writes one plain-text post draft per invocation to admin_docs/linkedin_posts/post_NN_<slug>.md, and updates the plan's status column. It never publishes anything (no LinkedIn API — the maintainer copies the text out and posts manually), never invents numbers or features not verifiable in the public repo, and never mentions students, grades, PII, secrets, or gitignored private content. Do NOT use this skill to write feedback or grades (separate skills exist), to e

Use this skill to evaluate one lecture folder in this course repo against the 9-category pedagogical rubric (six internal + three external) and produce a Markdown report with a composite tier and prioritized fixes. Trigger on requests like "evaluate lecture NN", "audit lecture NN", "score lecture NN", "review the lecture material for NN", or when the user supplies a `lectures_*/` folder path and asks for a quality assessment. Do NOT use it to: improve, rewrite, or propose content for a lecture (that work belongs to the future `create-improve-lecture` skill — this skill flags absence only); evaluate non-lecture material such as `README.md`, `CLAUDE.md`, or arbitrary scripts; grade student submissions or final-assignment notebooks (use `final_assignment/grade_feedback.prompt.md` instead). If the lecture number or path doesn't resolve to exactly one folder, ask which folder before proceeding.

Use this skill to produce a structural outline for one lecture in this course repo — a draft `goals_NN.md` plus an internal design doc that pre-conforms to the 9-category rubric used by `uoa-py-course-lecture-eval`. Trigger on requests like "outline lecture NN", "plan a lecture on X", "design lecture NN before writing it", "draft goals_NN.md for X", "plan the refactor of lecture NN", or "scaffold the outline for the regression lecture". Do NOT use it to: generate lecture content (no code, no datasets, no specific industry uses — that work belongs to the future `create-improve-lecture` skill); evaluate an existing lecture (use `uoa-py-course-lecture-eval`); grade student submissions; rewrite `docs/Lectures_outline.md` directly (the skill emits a suggested patch instead). If the lecture number or topic is ambiguous, ask before proceeding.

Use this skill when working inside this MLflow teaching repository to improve a tutorial notebook under `src/` (the `basics/`, `ml/`, or `gen_ai/` track) — expand, annotate, rewrite, draft a new section, restructure for flow, or clean up after a previous refactor — so it is friendlier for beginner data scientists and researchers. Trigger on prompts like "expand this MLflow section", "explain this MLflow concept better", "make this notebook beginner-friendly", "what context should we add to this cell", "why does this MLflow feature exist", "the logical flow here is broken", or "clean up the leftover trees / placeholder cells / old references". If the target notebook or cell is unclear, ask which one to edit after routing. Do NOT use it for general Python or ML tutorial writing outside this repo, for editing the upstream MLflow library itself, for non-tutorial docs in this repo (`README.md`, `CLAUDE.md`), or for non-teaching changes to the notebooks (file moves, dependency bumps, formatting, kernel metadata). F