Opinionated, lightweight agent skills for pretty much anything.
Start with the skills you need. Setup is optional: useful defaults are included in each skill, and setup records your project's preferences in AGENTS.md.
- Match the amount of planning, questioning, verification, and documentation to the task.
- Prefer the smallest clear, complete change. Add complexity when it has a concrete benefit, and tests when they provide meaningful confidence.
- Document current behavior and test current requirements. Retain implementation history when it serves a concrete ongoing need.
- Follow the user's request and existing project conventions. Use anything's defaults for choices they leave open.
- Delegate only when it is useful, available, and permitted. Work directly when it is not.
- Keep decisions and findings grounded in evidence. Distinguish independent review from self-review.
- Respect authorization already given within its scope. Ask when an action needs permission that is missing or unclear.
- Defer unrelated findings. Ask before pursuing them or adding them to an existing tracker unless already authorized; otherwise report them for later.
- Prefer Conventional Commits when Git is used and the project has no other convention.
The skills are instructions loaded when relevant or explicitly invoked; installing the plugin does not replace the host's instructions or grant additional permissions.
| Skill | Use it to |
|---|---|
| setup | Explicitly initialize or update project preferences in AGENTS.md. Manual invocation only. |
| cleanup | Explicitly sweep a repository for maintainability improvements, fix clear issues, and discuss larger changes. Manual invocation only. |
| interview | Clarify a topic, goal, decision, or set of requirements through an adaptive conversation. |
| orchestrate | Coordinate work at the right scale, with optional delegation. |
| task-plan | Keep a concise plan and handoff record when the task benefits from one. |
| fresh-review | Review an artifact using a separate context when possible, or an explicitly labeled self-review. |
| approval-gate | Resolve authorization for consequential actions without repeating approvals unnecessarily. |
| git-style | Prepare focused changes and use Conventional Commits by default. |
| repo-style | Keep project structure and documentation clear and proportional. |
| no-tombstones | Keep code, tests, comments, and documentation focused on current requirements during ordinary edits, review, and cleanup. |
Each skill works on its own. References to other skills are optional helpers; their installation is not a prerequisite.
Explicitly invoke setup when you want to configure a project. It reads what is already there, asks only about unresolved choices, and creates or updates a small section of the root AGENTS.md. Existing instructions are preserved. Without file access it returns a draft instead.
Project preferences persist; tool availability and permissions are checked in the current session. Ordinary tasks do not trigger setup.
Explicitly invoke cleanup for a full or targeted repository sweep covering duplication, complexity, structure, tests, documentation, and evidenced performance costs. It fixes clear issues that preserve behavior and discusses larger changes. Request findings only when you want an audit without edits.
The sweep can use subagents when useful, available, and permitted, or work directly. See repository cleanup for scope, examples, and how it composes with the other skills.
/plugin install anything --marketplace ronedgecomb/anything
codex plugin marketplace add ronedgecomb/anything
codex plugin add anything@anything-marketplacepi install git:github.com/ronedgecomb/anythingnpx skills add ronedgecomb/anythingChoose the skills and agents you want. See installation for local development, defaults for project preferences, and compatibility for version-specific details. No runtime dependencies are required by the instruction-only skills themselves.
skills/<name>/SKILL.md: portable behavioral instructions.- Skill-local
references/andassets/: optional briefs, examples, and templates. plugin.jsonand.claude-plugin/plugin.json: plugin manifests..agents/plugins/marketplace.jsonand.claude-plugin/marketplace.json: installation catalogs.package.json: Pi package metadata.assets/: wordmark and plugin icon.
Read AGENTS.md for repository conventions and verification guidance. CLAUDE.md imports the same instructions. These files guide work on this repository; installed skills do not inject them into your project.
Copyright 2026 Ron Edgecomb. Licensed under the Apache License, Version 2.0. Each skill bundles its license and attribution for individual installation.