Back to directory

Forge

An open-source runtime for Anthropic's Agent Skills standard that compiles a SKILL.md into a portable, hardened agent you can deploy on Kubernetes, on-prem, air-gapped, in CI, or as an A2A endpoint.

Forge

Forge compiles a SKILL.md file into a secure, runnable agent with egress controls, encrypted secrets, and audit logging, meant to run next to a service on infrastructure you already operate rather than as a hosted product. A setup wizard configures the provider, keys, channels, and skills, and a try command scaffolds a keyless demo agent using whatever model credential is already available, dropping you into a chat where every tool call and egress check is shown inline.

By default it makes only outbound connections, enforces egress allowlists, encrypts secrets, and opens no public listeners, while structured NDJSON audit logs record every action with correlation IDs. The same compiled agent can run locally, in Docker, or in Kubernetes. Its feature set includes SKILL.md-based agent definitions with YAML frontmatter, build- and runtime domain allowlists enforced through a subprocess proxy, outbound-only Slack and Telegram channel connectors, cron scheduling with channel delivery, session persistence with long-term vector-search memory, automatic multi-provider LLM failover, an MCP client for connecting to HTTP MCP servers, and a browser-based web dashboard.

More in Agent frameworks

See all 902
  • Recurse

    recurse.run

    A serverless harness that lets a coding agent design, test, and deploy its own specialist sub-agents as tools, MCP servers, or bots, defined by a schema-validated manifest.

  • ProofForge

    shulgin.is-a.dev

    An AI-agent pipeline that decomposes a mathematics problem, proves each piece, and formalizes the result in Lean 4 and Mathlib until the Lean kernel accepts it, with six of its proofs already merged into Google DeepMind's formal-conjectures repository.

  • local-coder

    github.com

    Builds a local coding-agent team on top of Ollama and OpenCode by detecting the machine's hardware, recommending models for seven specialist roles, and testing whether each assigned model can actually use tools, edit a repository and pass verification.