Menu
Create, monitor, message, and debug agents via the CLI. Use when creating agents, checking agent status, sending messages to running agents, attaching to tmux sessions, streaming agent output, or managing tool policies and approvals.
Installer avec Codex ou Claude Copiez ce prompt, collez-le dans Codex, Claude ou un autre assistant, puis laissez-le vérifier la page du skill et l'installer pour vous.
Basé sur la classification professionnelle SOC
Store, search, and manage shared knowledge through the memory service. Use for persisting information across agent sessions, semantic search over stored knowledge, and inter-agent knowledge sharing.
Browse, read, and author per-project knowledgebase documents through the knowledge service CLI. Use for creating, updating, listing, and reading markdown documents stored against a project, and reconciling DB rows against disk.
Cut a new agentd release. Suggests the next semver version from conventional commits (or accepts an explicit `x.y.z`), bumps the workspace version, regenerates the changelog with git-cliff, then commits, tags, and pushes the `v*` tag that triggers the Release workflow. Use when the user runs /release or asks to publish/cut/tag a new version.
Create a new GitHub plan issue with the "Plan: <title>" format, appropriate labels, and structured body. Use when the user wants to capture a high-level plan for future work.
Ask the human user a question during workflow execution and check for answers. Use for agent-to-human Q&A, collecting decisions, and reacting to human responses via ask_response workflow triggers.
Branch stacking and PR management with git-spice for the agentd project. Use when creating stacked branches, submitting PRs, syncing with upstream, resolving rebase conflicts, or navigating a branch stack.
| name | agent-ops |
| description | Create, monitor, message, and debug agents via the CLI. Use when creating agents, checking agent status, sending messages to running agents, attaching to tmux sessions, streaming agent output, or managing tool policies and approvals. |
Operational skill for managing agents through the agentd CLI.
# Basic agent creation
agent orchestrator create-agent \
--name my-agent \
--working-dir /path/to/project \
--model claude-sonnet-4-6
# With system prompt
agent orchestrator create-agent \
--name my-agent \
--working-dir . \
--model claude-sonnet-4-6 \
--system-prompt "You are a code review agent."
# With initial prompt (starts working immediately)
agent orchestrator create-agent \
--name my-agent \
--working-dir . \
--model opus \
--prompt "Review the latest PR"
# With worktree isolation
agent orchestrator create-agent \
--name my-agent \
--working-dir . \
--worktree \
--model sonnet
# JSON output for scripting
agent orchestrator create-agent --name test --working-dir . --model sonnet --json
Model aliases: sonnet → claude-sonnet-4-6, opus → claude-opus-4-6, haiku → claude-haiku-4-5-20251001
# List all agents
agent orchestrator list-agents
# Filter by status
agent orchestrator list-agents --status running
agent orchestrator list-agents --status pending
agent orchestrator list-agents --status stopped
agent orchestrator list-agents --status failed
# Get specific agent details
agent orchestrator get-agent <agent-id>
# JSON output
agent orchestrator list-agents --json
# Send a message/prompt to an agent
agent orchestrator send-message <agent-id> "Please review this file"
# Attach to agent's tmux session (interactive)
agent orchestrator attach <agent-id>
# Stream agent output (real-time monitoring)
agent orchestrator stream <agent-id>
# Stream all agents
agent orchestrator stream --all
# View current policy
agent orchestrator get-policy <agent-id>
# Set policy: allow all tools (default)
agent orchestrator set-policy <agent-id> '{"type": "AllowAll"}'
# Set policy: deny all tools (text-only)
agent orchestrator set-policy <agent-id> '{"type": "DenyAll"}'
# Set policy: only allow specific tools
agent orchestrator set-policy <agent-id> '{"type": "AllowList", "tools": ["Read", "Grep", "Glob"]}'
# Set policy: allow all except specific tools
agent orchestrator set-policy <agent-id> '{"type": "DenyList", "tools": ["Bash", "Write"]}'
# Set policy: require human approval for every tool
agent orchestrator set-policy <agent-id> '{"type": "RequireApproval"}'
When an agent has RequireApproval policy:
# List pending approvals
agent orchestrator list-approvals
agent orchestrator list-approvals --agent-id <agent-id>
# Approve a tool request
agent orchestrator approve <approval-id>
# Deny a tool request
agent orchestrator deny <approval-id>
Approvals timeout after 5 minutes by default.
agent orchestrator model <agent-id> sonnet
agent orchestrator model <agent-id> opus
agent orchestrator model <agent-id> claude-sonnet-4-6
# Delete a single agent
agent orchestrator delete-agent <agent-id>
# List then delete stopped agents
agent orchestrator list-agents --status stopped --json | jq -r '.[].id' | xargs -I{} agent orchestrator delete-agent {}
tmux -Vtmux list-sessionsagent orchestrator get-agent <id>agent orchestrator get-agent <id>tmux list-sessionstmux capture-pane -t <session-name> -p