gr agent

gr agent [command]

gr agent bump

Send "continue" to the agents an API error stopped (overloaded, a server error, a dropped connection)

Find this workspace's Claude Code agents whose last turn an Anthropic API error ended after the CLI's own retries (529 overloaded, a 5xx, a dropped or timed-out connection) and type "continue" into each, as send would. One line per agent: its id, what stopped it, its name, the CLI's words, and whether it was sent. Ids narrow it to those agents. --limited also bumps the agents a usage limit stopped, for once the limit has lifted or the account was swapped. --list only shows them. Exits non-zero when a send was refused.

gr agent bump [<id>...] [flags]

gr agent list

List agents

gr agent list [flags]

gr agent new

Start an agent with one complete brief

gr agent new <prompt | -> [flags]

gr agent read

Read an agent's conversation

gr agent read <id> [flags]

gr agent send

Type one line into an idle Claude Code agent's box and submit it (a message, or a slash command such as /login)

Type one line into a Claude Code agent's input box and press Enter, as its person would: a follow-up, or a slash command such as /login or /compact. The words are joined with spaces; - reads the line from stdin. Refused, with nothing typed, unless the agent is running, idle and showing its input box: never while it is in a turn or a question, permission prompt, menu or dialog is open, so it cannot answer one.

gr agent send <id> <text... | -> [flags]

gr agent stop

Stop and archive an agent, keeping its conversation (self: the agent running this command, which ends without going to the Inbox)

gr agent stop <id | self> [flags]

Every command also takes --root <folder>, the project to act on. Without it, gr uses the git repository you are in, or else the current folder.