oberskills plugin
A plugin of seven skills for Claude Code and Codex: prompts, subagents, skill authoring, editing, web search and a browser.
- 64 ★ on the repository
- Last commit
Extended access
Why this level
- The browser skill drives a live Chrome: clicks, forms, network interception and saved sessions.
- The skill-eval server launches real agent sessions and spends tokens.
Install
- Works in
Manual install
/plugin marketplace add ryanthedev/rtd-claude-innThen `/plugin install oberskills@rtd` and `/reload-plugins`. The MCP servers need `bun` on PATH.
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- The browser skill drives a live Chrome: clicks, forms, network interception and saved sessions.
- The skill-eval server launches real agent sessions and spends tokens.
This is third-party code. Review the repository files before installing.
What it does
oberskills collects skills for agents' weak spots. The `prompt` skill writes and audits prompts for Claude in DESIGN and REVIEW modes. The `agent` skill helps decide whether to hand work to a subagent and builds a four-part brief with model choice. The `skill-craft` skill drives skill creation and review with evaluation. The `write` skill edits and drafts prose at several edit depths, from proofreading to rewriting, and adapts to a chosen style. The `web-research` skill launches parallel search agents, and every finding carries a source address. The `clarify` skill clears up an ambiguous request with a series of questions before work starts. The plugin ships two MCP servers on Bun and TypeScript: `skill-eval` tests skills with live sessions, `mcp-browser` drives Chrome through 40 tools.
Who it is for. Authors of skills and prompts and those who build multi-agent chains in Claude Code or Codex and want to test skills by measurement rather than by eye.
Good fit when
- A system prompt must be audited like a hostile reviewer and return a verdict table.
- An orchestrator spawns subagents needlessly and you need a threshold for when it is justified.
- A new skill must be brought to working shape and you must confirm its description triggers on the right requests.
- Web search should give only real links, with no invented addresses.
- Prose needs editing at the right depth without turning proofreading into a rewrite.
Not a fit when
- You need one narrow skill: a whole seven-skill plugin adds surplus to the context.
- The machine has no `bun`: the `skill-eval` and `mcp-browser` servers will not start without it.
- The `write` skill is not meant for translation or code comments.
Example request
Review this system prompt as a reviewer and return a verdict table with fixes.Limitations
The `skill-eval` and `mcp-browser` servers need `bun` on PATH: dependencies install through a SessionStart hook into the plugin data directory. After install or update run `/reload-plugins` once. The `skill-eval` server launches real headless Claude sessions through the Agent SDK, so measurements spend tokens. The Codex version still uses the same Agent SDK, and a Codex-native eval runner does not exist yet. The `browser` skill drives a live Chrome, including forms and saved sessions, and needs careful access.
How to disable. Run `/plugin uninstall oberskills@rtd` in Claude Code. For Codex remove the plugin from the connected marketplace.
FAQ
How do I install oberskills in Claude Code?
Run `/plugin marketplace add ryanthedev/rtd-claude-inn`, then `/plugin install oberskills@rtd` and `/reload-plugins`. Updating uses `/plugin update oberskills@rtd`. `bun` on PATH is needed for the two MCP servers.
Which skills are in the plugin?
Seven: `prompt`, `agent`, `skill-craft`, `write`, `web-research`, `browser` and `clarify`. The three meta-skills `prompt`, `agent` and `skill-craft` were fully rewritten in version 2.0.0.
What does the skill-eval server do?
Seven tools: `validate_skill`, `test_triggers`, `optimize_description`, `run_eval`, `grade_run`, `aggregate_benchmark` and `compare_outputs`. They check skill structure and description triggering and compare outputs blind.
Does oberskills work in Codex?
Yes, the repository keeps one shared source tree for both hosts: Claude Code takes `.claude-plugin/plugin.json`, Codex takes `.codex-plugin/plugin.json` and `.mcp.json`. MCP dependencies install on first start.
What does the write skill do in EDIT and REVIEW modes?
EDIT rewrites silently. REVIEW walks through issues in batches, asks questions, then offers an edit. Depth runs from proofreading to rewriting, and style comes from Economist, AP and GOV.UK cards.
Discussion
No questions or reviews yet.
Related
PluginsA plugin with six working skills and a prompt library: routing a task to a prompt, pre-commit checks and skill audits.
A skill for creating skills that proves them by behavior: before-and-after runs, regression evals and ecosystem audit
The author's personal Claude Code toolbox: meta-prompt and todo commands, skills to create skills, hooks and MCP servers, auditors and a Ralph loop
A skill registry for teams with security scanning and versioning; the MCP server finds and installs skills on request.
A desktop app and CLI to install, sync and organize agent skills across Claude Code, Codex, Cursor and other tools
Claude Skills MCP: semantic search for skills
Claude Skills MCP
An MCP server with vector search: the agent describes a task and gets matching skills from the Anthropic and K-Dense collections
README in short
The author states the goal as making Claude Code and Codex better where they are weakest. Version 2.0.0 draws on a review of Anthropic documentation and about a hundred arXiv papers.