Use this skill whenever the user asks about unit test best practices, test organization, flat test suites, redundant tests, test refactors, pytest/JUnit/Jest/xUnit layout, test taxonomy, flaky tests, coverage quality, fixtures, mocks, parametrization, pruning existing UTs, or "which tests are worth keeping." It aggressively turns broad testing advice into a clean, behavior-first UT suite by preventing random test growth, pruning low-value tests, consolidating redundant checks, and requiring each retained test to prove a real behavior, failure mode, or public contract. For broad suite refactors, audit first, propose a recommended path across pruning, markers, folder layout, fixtures, and parameterization, then wait for user feedback before applying disruptive changes.
Initialize, audit, aggressively slim, merge, and refresh project-local AGENTS.md and CLAUDE.md files from existing repo guidance, agent /init suggestions, stdin-bundled Codex init-style discovery, and intuitive workflow defaults. Use when setting up a repo for Claude Code/Codex, replacing symlinked agent files with local guidance, rerunning agent init after weeks of drift, cleaning overgrown root agent files or first-read policies, optimizing startup context across AGENTS/CLAUDE/README/ARCHITECTURE/STATUS, adding worktree-friendly .venv/submodule hooks and hints when repos already show Python environment or git-submodule signals, or configuring target-repo LSP and agent-facing Serena MCP guidance. Route broad human docs, tests, execution flow, refactor scope, and entropy discovery to their own skills unless the change is the startup/orientation harness that points at them.
Track goal-driven Multica issues. Use when a user creates or maintains Multica issues by pasting a /goal prompt, wants Codex to summarize that goal into a concise issue purpose, append a normalized goal-start comment, record completion evidence from real Multica execution run messages or a supplied session transcript, or add additional goal attempts and completion proof to the same issue.
Run one approved implementation plan through multiple isolated candidate worktrees, each with its own agent harness/model/profile, then compare artifacts and rank the results. Use when the user asks for best-of-N plan execution, model or harness bakeoffs, parallel candidate worktrees, fast-model implementation with stronger-model judging, or safe evaluation of Codex, Claude Code, Kimi, MiniMax, or custom coding-agent routes against the same real project plan.
Stable execution/change router after an approved plan, preflight contract, or tiny concrete task. Refactor-shaped work delegates to intuitive-refactor, and durable work runs through staged planning, review, GSD handoff, implementation, cleanup, and verification while keeping plan ledgers and active capsules compact by replacing stale state instead of appending history. Also accept vague prompts as a compatibility route that names the upstream planning stage instead of hiding planning inside execution.
Periodically inspect a repository or plan and produce a ranked batch of high-value entropy reduction candidates. Use repo entropy mode when the user says the repo feels messy, asks what to clean next, wants to make an old repo easier for humans and AI agents, or wants maintenance suggestions without already knowing the target seam. Use plan entropy mode when the user points at an idea, draft plan, or named plan file and wants blind spots, missing decisions, or weak assumptions found before grill-batch/preflight. The skill must state the selected mode and why before auditing, return a no-change result instead of filling a requested count when only polish remains, choose an explicit discovery intensity, run saturation discovery when the user asks for all directions or unknown unknowns, and end with one recommended next action plus a short reply shortcut. For approved repo-wide maintenance goals, return a deduplicated maintenance handoff for intuitive-refactor.
Refactor and cleanup router for known code/module/API seams, stale surfaces, compatibility shims, architecture cleanup targets, changed-code quality review, oversized modules, repeated cleanup campaigns, and recurring whole-repo architecture maintenance goals. Use this when the user names a concrete seam, wants code/package layout made smaller and truer, asks to keep refactoring, or asks for a goal that periodically cleans architecture across the whole repo until saturated. For whole-repo maintenance goals, run the repo-wide maintenance loop: alternate intuitive-reduce-entropy discovery handoffs with execution of every clear safe candidate, deduplicate parked and low-value registries, and stop only when saturation finds no new clear P1/P2. For unknown "what should we clean?" discovery-only prompts, start with intuitive-reduce-entropy, then return here to execute selected candidates.
Create and maintain an intuitive human documentation surface for AI-agent-developed repos. Use when humans should only need README.md, ARCHITECTURE.md, STATUS.md, and docs/human/** while planning logs, generated docs, retrospectives, ADR detail, and implementation evidence stay in AI-agent-only folders.