agent-orchestrator
Manage the full agent lifecycle, workflows, approvals, tool policies, and runtime configuration through the orchestrator service.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
메뉴
Manage the full agent lifecycle, workflows, approvals, tool policies, and runtime configuration through the orchestrator service.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
SOC 직업 분류 기준
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-orchestrator |
| description | Manage the full agent lifecycle, workflows, approvals, tool policies, and runtime configuration through the orchestrator service. |
Skill for interacting with the agentd orchestrator service — the central coordinator for AI agents and autonomous workflows.
# Basic agent
agent orchestrator create-agent \
--name my-agent \
--working-dir /path/to/project \
--model claude-sonnet-4-6
# With an initial prompt (starts working immediately)
agent orchestrator create-agent \
--name my-agent \
--working-dir . \
--model opus \
--prompt "Review the latest PR"
# With system prompt for persistent instructions
agent orchestrator create-agent \
--name my-agent \
--working-dir . \
--model sonnet \
--system-prompt "You are a code review agent. Focus on security issues."
# With worktree isolation (creates a git worktree)
agent orchestrator create-agent \
--name my-agent \
--working-dir . \
--worktree \
--model sonnet
# With environment variables
agent orchestrator create-agent \
--name my-agent \
--working-dir . \
--model sonnet \
--env GITHUB_TOKEN=xxx \
--env RUST_LOG=debug
# Docker-backed agent with resource limits
agent orchestrator create-agent \
--name my-agent \
--working-dir . \
--model sonnet \
--docker-image agentd-runner:latest \
--cpu-limit 2.0 \
--memory-limit 4096
# Read prompt from stdin
echo "Fix the failing tests" | agent orchestrator create-agent \
--name fixer \
--working-dir . \
--model sonnet \
--stdin
# Auto-attach to the agent's tmux session after creation
agent orchestrator create-agent \
--name my-agent \
--working-dir . \
--model sonnet \
--attach
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 (pending, running, stopped, failed)
agent orchestrator list-agents --status running
# Get detailed info for a specific agent
agent orchestrator get-agent <agent-id>
# JSON output for scripting
agent orchestrator list-agents --json
# Send a message/prompt to a running agent
agent orchestrator send-message <agent-id> "Please review this file"
# Send a message from stdin
echo "Fix the bug in auth.rs" | agent orchestrator send-message <agent-id> --stdin
# Attach to agent's tmux session (interactive terminal)
agent orchestrator attach <agent-id>
agent orchestrator attach --name my-agent
# Stream real-time output via WebSocket
agent orchestrator stream <agent-id>
agent orchestrator stream <agent-id> --verbose
agent orchestrator stream --all
# View Docker agent logs
agent orchestrator logs <agent-id>
agent orchestrator logs <agent-id> --follow
agent orchestrator logs --name my-agent --tail 200
# Get usage statistics (tokens, cost, turns, duration)
agent orchestrator usage <agent-id>
agent orchestrator usage --name my-agent
# Clear context and start fresh
agent orchestrator clear-context <agent-id>
agent orchestrator clear-context --name my-agent
agent orchestrator set-model <agent-id> --model sonnet
agent orchestrator set-model --name my-agent --model opus
agent orchestrator set-model <agent-id> --model opus --restart
agent orchestrator set-model <agent-id> --clear # Remove model override
# Add an additional directory to an agent's accessible paths
agent orchestrator add-dir <agent-id> /path/to/extra/dir
# Remove an additional directory
agent orchestrator remove-dir <agent-id> /path/to/extra/dir
agent orchestrator delete-agent <agent-id>
# Batch delete stopped agents
agent orchestrator list-agents --status stopped --json | jq -r '.[].id' | xargs -I{} agent orchestrator delete-agent {}
Control what tools an agent can use.
# View current policy
agent orchestrator get-policy <agent-id>
# Allow all tools (default)
agent orchestrator set-policy <agent-id> '{"type": "AllowAll"}'
# Deny all tools (text-only mode)
agent orchestrator set-policy <agent-id> '{"type": "DenyAll"}'
# Only allow specific tools
agent orchestrator set-policy <agent-id> '{"type": "AllowList", "tools": ["Read", "Grep", "Glob"]}'
# Allow all except specific tools
agent orchestrator set-policy <agent-id> '{"type": "DenyList", "tools": ["Bash", "Write"]}'
# Require human approval for every tool call
agent orchestrator set-policy <agent-id> '{"type": "RequireApproval"}'
When an agent has RequireApproval policy, tool calls queue for human review.
# List pending approvals
agent orchestrator list-approvals
agent orchestrator list-approvals --agent-id <agent-id>
agent orchestrator list-approvals --status pending
# Approve or deny
agent orchestrator approve <approval-id>
agent orchestrator deny <approval-id>
Approvals time out after 5 minutes by default.
Workflows automate task dispatch to agents from external sources.
# GitHub issues workflow
agent orchestrator create-workflow \
--name issue-worker \
--agent-name worker \
--trigger-type github-issues \
--owner geoffjay \
--repo agentd \
--labels agent \
--state open \
--poll-interval 60 \
--prompt-template 'Work on issue #{{source_id}}: {{title}}\n\n{{body}}'
# GitHub pull requests workflow
agent orchestrator create-workflow \
--name pr-reviewer \
--agent-name reviewer \
--trigger-type github-pull-requests \
--owner geoffjay \
--repo agentd \
--labels review \
--state open
# Cron-based workflow
agent orchestrator create-workflow \
--name nightly-check \
--agent-name checker \
--trigger-type cron \
--cron-expression "0 2 * * *"
# Webhook-triggered workflow
agent orchestrator create-workflow \
--name deploy-hook \
--agent-name deployer \
--trigger-type webhook \
--webhook-secret mysecret
# Manual trigger only
agent orchestrator create-workflow \
--name on-demand \
--agent-name worker \
--trigger-type manual
Template variables: {{source_id}}, {{title}}, {{body}}, {{url}}, {{labels}}
agent orchestrator list-workflows
agent orchestrator get-workflow <workflow-id>
agent orchestrator update-workflow <workflow-id> --enabled true
agent orchestrator update-workflow <workflow-id> --enabled false
agent orchestrator update-workflow <workflow-id> --poll-interval 120
agent orchestrator delete-workflow <workflow-id>
# View dispatch history
agent orchestrator workflow-history <workflow-id>
# Manually trigger a workflow
agent orchestrator trigger-workflow <workflow-id> --title "Manual task" --body "Do this thing"
# Validate a prompt template
agent orchestrator validate-template '{{source_id}} - {{title}}: {{body}}'
agent orchestrator validate-template --file template.txt
agent orchestrator health