Verified Agent Skill record

Autogpt Agents

Skill guide for the AutoGPT autonomous AI agent platform: building agents as graphs of blocks in the visual builder, running continuous executions with manual, webhook, and scheduled triggers, creating custom agents with the Forge toolkit, and benchmarking with agbenchmark. Covers platform setup via Docker and npm, integrations with providers such as OpenAI, GitHub, and Notion, and monitoring executions via WebSocket and REST APIs.

Tasks

Build a visual workflow agent in the AutoGPT Platform builder using blocks connected as a graph
Create a Forge-based custom agent with a custom ability under autogpt_platform/classic using ./run forge create and ./run forge start
Set up webhook or scheduled triggers for agent execution via the REST API
Run benchmarks with ./run benchmark, including specific categories or agents

Inputs

Agent graph definitions composed of nodes, links, and blocks
Environment variables (.env) including DATABASE_URL, REDIS_URL, RABBITMQ_URL, ENCRYPTION_KEY, SUPABASE_URL
Inputs to graph executions (e.g., JSON payloads for manual, webhook, or scheduled triggers)
API keys or OAuth authorizations for providers such as OpenAI, Anthropic, GitHub, Google, Discord, Notion

Outputs

Running agent executions producing output yields from AgentOutputBlock nodes
Execution status updates via WebSocket (ws://localhost:8001/ws) or REST polling (GET /api/v1/executions/{execution_id})
Benchmark results across coding, retrieval, web, and writing categories

Limitations and checks

The skill metadata lists an MIT license, but the source document itself states the Platform is under Polyform Shield while Classic is MIT; the exact licensing for the platform component is not resolved in the source
No verified skill-specific install commands exist for this skill; installation guidance consists only of general AutoGPT platform setup steps from the source document
Some references (references/advanced-usage.md, references/troubleshooting.md) are mentioned in the source but not included in the skill evidence
Credential security claims ('encrypted and stored securely') are asserted by the source document and are not independently verified
Confirm Docker, npm, and Poetry are available and required environment variables (DATABASE_URL, REDIS_URL, RABBITMQ_URL, ENCRYPTION_KEY, SUPABASE_URL) are set before starting services
Verify the frontend (http://localhost:3000), backend API (http://localhost:8006/api), and WebSocket (ws://localhost:8001/ws) endpoints are reachable after docker compose up and npm run dev
Run ./run benchmark against a created Forge agent to confirm agent quality before deployment
If services fail, check docker compose ps, docker compose logs rest_server, run poetry run prisma migrate deploy for database issues, and check the RabbitMQ queue at http://localhost:15672 for stuck executions