rmkr-skills
My agent skills for planning, execution, review, and communication across subjects. Built for experienced agent users, with Codex as the primary runtime and best-effort support elsewhere.
Choose a skill
| Skill | Use it when | Required companions | Invocation |
|---|---|---|---|
| ArtCraft | Create or edit a file with an ArtCraft app, such as combining PDFs or editing a named vector object. | The selected app’s CLI or connected interface | Automatic or explicit |
| Subagent Delegation | Parts of a task benefit from separate agents, with one main agent coordinating the results. No formal plan or implementation worker is required. | Optionally Setup rmkr skills to configure model tiers | Automatic or explicit |
| forge | Work needs independent scrutiny of its assumptions and delivery, across software, research, plans, documents, and other subjects. Levels: review returns verified findings, fix corrects and re-reviews, and build has workers implement software first. Loops stop when clean or at the shared round limit; unfinished work stays explicit. Each finding is checked and fixed on the lowest model tier that fits it. Add strict for a runtime-enforced read-only reviewer on a Git diff. | Subagent Delegation; the strict reviewer for strict; optionally Setup rmkr skills to configure its reviewer policy | Automatic or explicit |
| Unslop | Prose needs formulaic phrasing removed while preserving meaning and evidence. | None | Automatic or explicit |
| Setup rmkr skills | You want global or per-repository settings for subagent model tiers, forge's reviewer policy, and its review-round limit. | None | Explicit only |
A review request does not authorize edits. Reviewers inspect; verifier subagents check findings, and the main agent makes the final call on each verdict and owns any separately authorized corrections. forge's strict mode requires runtime-enforced read-only and no-delegation controls. Routine independent review uses ordinary native subagents with inspection-only assignments.
Installation
For Codex CLI, install the collection as one plugin from the development branch:
codex plugin marketplace add rmkr/skills --ref main
codex plugin add rmkr-skills@rmkr-skills
Start a fresh session. Plugin skills use a prefix, for example $rmkr-skills:artcraft. See installation for updates, a daily macOS update task, and migration from an existing skill installation. These commands require the plugin files on main; until this change is merged, use the local test instructions in that guide.
For Claude Code, install the same collection as a plugin:
claude plugin marketplace add rmkr/skills#main
claude plugin install rmkr-skills@rmkr-skills
Start a fresh session and use commands such as /rmkr-skills:artcraft. The strict reviewer still needs separate setup. These commands require the Claude plugin files on main; until merged, use the local test instructions.
For individual skills, choose one command:
| Tool | Command |
|---|---|
| npm | npx skills@latest add rmkr/skills --global |
| Bun | bunx skills@latest add rmkr/skills --global |
Choose the skills and agents when prompted. Keep Symlink selected if asked for an installation method. This lets agents share one installed copy of each skill.
Install subagent-delegation with forge. Start a fresh agent session after installation.
See installation for forge's optional strict reviewer setup, skills CLI documentation for other commands, and the changelog for collection changes.
Skills installed through the skills CLI use $skill-name in Codex, for example $unslop. See compatibility for other runtimes, invocation policies, and verification limits.
Contribute and maintain
Contributions are welcome, including new skills within the collection's scope. Small fixes can go straight to a pull request; discuss new skills and major behavior changes in an issue first. The contribution guide explains acceptance criteria, checks, review, and collection-wide releases. The maintainer retains scope, merge, and release decisions. Maintenance is best effort, without promised response times or a release schedule.
Recommendations and license
Personal recommendations cover third-party skills, plugins, and MCPs. Those projects retain their own maintainers and support channels.
Project-authored material is available under the MIT license. Existing third-party license notices and attribution remain applicable to their material.