| name | deploy-managed-agents |
| description | Deploy, run, list, or tear down the agents in this repo's .managed-agents/ folder on Anthropic Managed Agents. Use when the user says "deploy my agents", "ship this agent to the cloud", "run the deployed agent", or asks about managed-agent deployment. |
Deploy managed agents (agentlift)
This repo defines deployable agents under .managed-agents/. Use the agentlift
CLI to push them to Anthropic's Managed Agents cloud and run them by ID. The user
shouldn't need to learn it — translate their intent into the right command.
Prerequisite: pip install agentlift and ANTHROPIC_API_KEY set (it's in .env).
What the user wants -> what to run
- "show me what would deploy" / "dry run" ->
agentlift plan .
- "is this valid?" ->
agentlift validate . (exit 1 means there are errors to fix)
- "deploy" / "ship to the cloud" ->
agentlift deploy . --yes
- then report the agent IDs printed, and that
.agentlift-lock.json was written.
- "redeploy" / "I changed an agent" ->
agentlift deploy . --yes --prune
- idempotent: unchanged skills/agents are skipped automatically.
- "run with " ->
agentlift run <agent> --project . --task "<task>"
- "run it locally" / "same agent on my machine" -> add
--local
- "what's deployed?" ->
agentlift list .
- "tear it down" / "delete the agents" ->
agentlift destroy . --yes
Rules
- Always run
agentlift plan . first and show the user the plan before deploy.
Surface any ERROR diagnostics (e.g. a stdio MCP server, which can't deploy)
and stop — don't pass --skip-unsupported unless the user agrees to drop those.
- After deploying, the
.agentlift-lock.json maps each agent to its remote ID.
Tell the user to commit it so re-deploys are idempotent and run works for the
whole team.
- Never put
ANTHROPIC_API_KEY on the command line or in a committed file.
