python-refactor

Explore a Python codebase to find opportunities to deepen shallow modules, reduce coupling, and improve testability. Reads codebase_metrics.json from .python-refactor/ (auto-runs measure if missing) and cross-references BUG_REPORT_<RUN_ID>.md if present to show which architectural patterns are hiding confirmed bugs. Presents numbered candidates, spawns parallel design subagents for the chosen candidate, recommends the strongest interface design, and creates issue RFCs. Use when the user wants architecture review, module deepening, or coupling reduction.

Remove all python-refactor plugin output and project-root tool caches. Deletes the entire .python-refactor/ directory (all runs, all deliverables) and project-root tool caches (.hypothesis/, .pytest_cache/, .semgrep/, .skylos_cache/) unconditionally. This is a hard nuke — use when you are done with all analysis and want a clean project root. Safe to run multiple times. Does not touch global tool caches or filed issues.

Actively hunt real bugs in a Python codebase using two phases. Phase 1 is free agentic organic exploration guided by static metrics — the agent reads high-signal files completely, reasons step-by-step through logic errors, Python footguns, semantic drift, and integration failures, and builds a candidate list. Phase 2 uses Hypothesis property-based testing to confirm and reproduce each candidate. Outputs BUG_REPORT_<RUN_ID>.md and Hypothesis test files to .python-refactor/. Auto-runs python-refactor:measure if codebase_metrics.json is missing. When invoked by the orchestrator, returns a compact JSON handoff.

Instrument a Python project with static analysis tools and write codebase_metrics.json to .python-refactor/. Runs radon (cyclomatic complexity, maintainability index, Halstead metrics), vulture (dead code), bandit (security patterns), semgrep (OWASP), skylos (taint analysis), and grimp (module coupling graph) in parallel. Raw tool output goes to .python-refactor/tmp/ and is cleaned after aggregation. Invoked automatically by other skills in this suite when codebase_metrics.json is missing. When invoked by the orchestrator, returns a compact JSON handoff instead of a prose summary.

Full python-refactor pipeline orchestrator. Chains measure, hunt-bugs, architecture, and plan-refactor in strict sequence using isolated subagents. Each phase receives only a compact JSON handoff from the previous phase — never the full session history. This prevents context rot across a long pipeline. After all phases complete, files a single summary issue with collapsible detail sections, then cleans up all temporary and deliverable files for this run. All output goes to .python-refactor/ at the project root. Use when the user says "analyze this codebase", "full refactor analysis", "find all issues", or invokes /python-refactor:orchestrate directly.