Create, improve and manage Claude Code agents without editing Markdown by hand

The problem

A Claude Code agent is a Markdown file with a name, a description of when to use it, the tools it may use and its instructions. They're powerful, but they're scattered: some in ~/.claude/agents, some in each repo's .claude/agents, some shipped inside plugins. It's hard to remember which ones exist, which repo they belong to, and whether anyone actually uses them.

Writing a good one is also slow. The best instructions come from real chats where you corrected the agent, and turning those corrections into rules by hand is the step most people skip.

How Pairbar solves it

1. Open the Agents tab
   Each agent shows its repo, whether it's read-only or edits code, its model, how many commands it has, and an orange new or edited tag when its file isn't committed yet.
2. Create one with New agent
   Click New agent and describe it, e.g. "a reviewer that checks database migrations for missing indexes and down steps". A composer opens with the job attached; add a line and send. The Claude chat writes the agent, adds its commands and installs it.
3. Fix what it keeps getting wrong with Improve
   On an agent's page, Improve takes a note like "keeps missing migrations" and Claude edits that agent.
4. Learn from real chats
   From a chat's menu, Agents › Turn this chat into an agent makes a reusable agent from a chat that went well, focusing on what you corrected. Teach an agent from this chat adds concrete rules to an existing agent from where it went wrong.
5. Use it
   Start it from its page (Start runs a ready-made command, Blank chat starts empty), from the composer's Agent chip, from a chat's Agents › menu, or with ⌘K → "Start ".
6. Share it with Codex
   Add to Codex exports the agent to ~/.codex/agents/<name>.toml so Codex can use it too. It's on the agent's page, next to In Claude ✓.

Where agents come from

An agent's page

• A one-line summary with More about it, plus repo, model, access and tools.
• Activity: how often it ran in the last 30 days and the chats it ran in, from Claude's own record of subagent runs.
• Instructions, collapsed by default.
• ⋯ menu: Edit manually, Duplicate, Copy @mention, Open file in…, Reveal in Finder, Move to Trash.

Edit manually is a form for quick tweaks: name, title, description, model, group, repo and icon; tools (without Edit and Write the agent is read-only); commands with {input} placeholders; and the instructions. Saving an agent you haven't changed leaves its files byte-for-byte identical.

To give agents groups, icons, a starting repo and ready-made commands, add an optional dock.yaml next to your agent files. See organise agents with dock.yaml. New agents are saved where your agents already live; change that in Settings → Agents.

Without Pairbar

Claude Code has its own ways to manage agents:

• The /agents command inside Claude Code lists agents and walks you through creating or editing one.
• Write the Markdown yourself with YAML frontmatter (name, description, tools, model) in ~/.claude/agents/ or .claude/agents/, following Anthropic's documented format.
• Commit project agents to the repo so the whole team shares them through git.