Orchestrate herdr terminal sessions, spaces, tabs, and coding agents via the herdr CLI.
npx skills add https://github.com/jlui17/dotfiles --skill herdr-agents-jlui17
Please help me install this Agent Skill. Skill: herdr-agents Source: https://github.com/jlui17/dotfiles/tree/main/agents/skills/herdr-agents Command: npx skills add https://github.com/jlui17/dotfiles --skill herdr-agents-jlui17
--session, --workspace, --cwd, and --label targeting so work lands in the right place. - Agent lifecycle management: Start named agents of any kind (claude, codex, gemini, opencode, amp) with a launch-time kickoff prompt, send follow-up prompts, and block on completion with agent wait or prompt --wait. - State monitoring and verification: Read semantic agent states (working, blocked, idle, done) via agent list, inspect actual terminal output with agent read, and verify external artifacts directly instead of trusting idle status. - Use Case: Kick off a new task by creating a worktree-backed space, spawning a Claude agent in its own tab with a one-line prompt pointing at a brief file, waiting for it to settle, reading its output, and closing the tab when done. ## Quick Start Ask the agent to start a new herdr session named "review", create a workspace in your project directory, spawn a Claude agent with a kickoff prompt, and report back when it finishes.High-intent search queries and answers about installing and using this skill.
Create a tab with herdr tab create --workspace <id> --cwd <dir> --label <task>, then run herdr agent start <name> --kind claude --pane <pane_id> -- "<prompt>". Pass the kickoff prompt at launch, since prompting a fresh agent afterward can silently fail.
Use herdr agent wait <name> --until blocked --until done --timeout <ms> to block until the agent settles, or herdr agent prompt <name> "<prompt>" --wait to submit and block in one call. Check herdr agent list for a fleet-wide view of agent states.
A prompt sent to a non-working agent must produce an observed lifecycle change within five seconds or herdr returns agent_prompt_stalled. Right after agent start this is usually a startup race where the submit did not land; retry once and confirm with agent read.
No, a launch-argv prompt must be one line or it fails with invalid_agent_argument. Write the full brief to a file and pass a one-liner telling the agent to read and follow that file.
Not necessarily. An idle agent's own watchers, such as PR monitors or poll loops, can die silently. Verify the external artifact directly, like the PR's reviews or the job's state, instead of trusting agent list idle status.
Run herdr tab close <tab_id> to kill a tab and its processes, then herdr session stop <name> followed by herdr session delete <name> for sessions. If closing your own tab, do it last since it kills your own process.
dushyantkhoslaCreate isolated Git worktree workspaces for feature development.
StartaleGroupAnalyze pull request diffs to assign review tiers and agents.
mkalkerePublish and consume structured JSONL messages on a file-based bus.
prismatic-koiSpawn isolated agent sessions in git worktrees via tmux.