Skills (F4)

Skills are short instructions an agent reads when a task calls for them. nerdr installs its own, so that the agents know how to use nerdr: how to explain a change, ask for a review or message another agent. You do not call them yourself; you ask in plain words, and the agent picks the skill.

nerdr’s skills

Skill	What it lets the agent do	Described in
nerdr-explain	Explain a finished change, or walk through it, when asked	Code changes
nerdr-request-review	Ask for a code review when the work is done	Code review
nerdr-send	Message the other agents in the tab, start more, and follow them	Agents working together
nerdr-spinoff	Split work into new tabs	Splitting work
nerdr-handoff	Write down its work for the agent taking over	Switching agents
nerdr-tools	Keep a line on where its work stands, see every tab, run commands in a terminal you see, make tabs, search past conversations, check rate limits	Commands for agents
nerdr-plugins	Find and run your plugins	Plugins

Their sources are in skills/. A skill is a folder: its SKILL.md, and sometimes files beside it that the agent reads only when it needs them, such as the walkthrough part of nerdr-explain.

The F4 screen

F4 has three lists. nerdr shows nerdr’s skills and whether the tab’s agent has them: ✓ installed, ~ an older copy that is updated at the next start, ✗ missing. other shows the skills in the folders the tab’s agent reads, and any others used, such as an agent’s built-in or plugin skills; nerdr can only read these, and does not turn them on or off. Both are sorted by name. context shows what else the agent reads at start: instruction files, rules, memory, MCP servers, and the note nerdr adds itself. These are the places the agent’s documentation names, narrowed to what exists; for Cursor and Antigravity they are the usual places. The list is read-only, shows only the names of MCP servers, and leaves out servers from plugins and connectors.

Key	Action
tab	Switch between the lists
a	Count use in this tab or in all tabs
↑ / ↓	Read a skill or file
space	Turn one of nerdr’s skills on or off
pgdown / b	Scroll
s	Change where skills are installed

F4 also counts how often each agent has used each skill and when it last did, in the active tab until you press a. A conversation joins a tab’s count when nerdr counts it there, so older conversations may show only under all tabs. nerdr counts from each agent’s conversation log when a turn ends. These are use counts, not measured token costs. One of nerdr’s skills shows Without this skill first: CLI examples, what the skill adds, and what turning it off changes.

Turning skills off

All skills start on. Their descriptions can take context before use; reading their instructions, calling tools and doing the work also cost tokens. Manual CLI use can avoid agent calls, but asking an agent to read CLI help and do the same work still costs tokens. Reviews and new agents still use tokens when started through the CLI.

Turn off skills you do not use in F4 or with enabled_skills:

enabled_skills = ["nerdr-explain", "nerdr-spinoff"]

An empty list turns all of them off. Claude notices a change by itself; restart other running agents. Turning a skill off cannot remove text already in a conversation. Your own copies of skills remain discoverable until you remove them yourself.

F1 explanations and walkthroughs, and handoffs requested when switching agents, work with their skills off. nerdr includes the necessary instructions in those requests, only when you use the action. Keep nerdr-explain and nerdr-handoff on if you want the agent to discover those workflows from ordinary conversation too.

Turning off nerdr-request-review also changes automatic review: with when = "ready", nerdr falls back to reviewing each turn instead of waiting for the agent to say it is ready. Turn automatic review off separately if you want to avoid those reviews.

CLI alternatives

Use nerdr COMMAND -h for options and input formats. Replace uppercase placeholders with your values. Commands controlling tabs need the agent’s nerdr environment, including its connection to the running tab. Terminal panes do not have that connection either: ask the agent to run the exact command without loading a skill. Local note commands (order, why, walkthrough) and plugin commands can also run from a shell. Agents should use "${NERDR_BIN:-nerdr}" instead of nerdr. See Commands.

Skill	Commands without the skill	What you supply or lose
nerdr-tools	nerdr status "Working", nerdr tabs, nerdr terminal npm run dev, nerdr terminal --restart terminal, nerdr read terminal, nerdr close terminal, nerdr tab new --branch fix/example "Task", nerdr search WORDS, nerdr search --show ID, nerdr usage	Individual actions still work. The agent loses guidance on status updates, visible long runs and finding past decisions. Sandbox restrictions still apply.
nerdr-send	nerdr agents, nerdr send 2 "Message", nerdr agent --kind codex "Task", nerdr wait --timeout 1m 2, nerdr read 2, nerdr close 2	Supply complete tasks and coordinate replies yourself. Starting and messaging agents may still cost tokens; closing a pane stops its work.
nerdr-plugins	nerdr plugin list, nerdr plugin run NAME --PARAM VALUE, nerdr plugin stop NAME	Choose the plugin and parameters yourself. Permissions and confirmations remain; plugins for the user only open in F5. The skill also teaches watch handling and plugin creation.
nerdr-request-review	nerdr request-review "Review focus", optionally --agent codex before the text	Request at the right time; the review starts after the agent’s turn. The skill teaches when to request and what to highlight. Automatic review fallback is described above.
nerdr-spinoff	nerdr spinoff plan.json	Prepare the split, commits and full task descriptions as JSON. The command opens the confirmation dialog; it does not work out the plan. The agent also loses the reminder to use spinoff for new branches. nerdr tab new creates a tab directly, without that dialog or splitting existing changes.
nerdr-handoff	nerdr handoff handoff.md	Write the summary yourself. The command sends it; it cannot recover decisions from the conversation. The agent-switch dialog supplies writing instructions when needed.
nerdr-explain	nerdr order add --short "Summary", nerdr order add --file FILE --role core --short "Role", nerdr why add --file FILE --anchor "Line" --short "Reason", nerdr walkthrough add walkthrough.json	Write the explanations and walkthrough structure yourself. These commands record content rather than generate it. E and g / G in F1 supply writing instructions when needed.

Where skills are installed

install says where nerdr may write what it gives the agents: the skills, and the hooks and plugins that tell nerdr what an agent is doing. You choose on first start.

With install = "home", nerdr installs for the agents found on your PATH:

Agent	Folder
Claude	~/.claude/skills/
opencode	~/.claude/skills/, or ~/.config/opencode/skills/ without Claude
Copilot	~/.copilot/skills/
Codex	~/.agents/skills/
Gemini	~/.agents/skills/
Antigravity	~/.gemini/config/skills/
Grok	~/.agents/skills/
Cursor	~/.agents/skills/ or ~/.claude/skills/; otherwise ~/.cursor/skills/

The agents then find the skills outside nerdr too.

With install = "worktree", skills go in the folder each agent starts in: .claude/skills/ for Claude, opencode, Copilot and Cursor, and .agents/skills/ for Codex, Gemini, Antigravity and Grok, kept out of git through .git/info/exclude. Use this for sandboxes that cannot reach your home folder. Grok reads a project’s skills only once you trust its folder.

With install = "off", nerdr installs nothing. Copy the folders from skills/ yourself for discovery through conversation. F1 explanations, walkthroughs and agent-switch handoffs still carry their own instructions.

nerdr updates its skills at every start. To keep your own edits to one, remove the <!-- nerdr: installed and kept up to date by nerdr … --> marker from the file; files without it are left alone. Turning a skill off removes nerdr’s copies, but keeps your own.