Documentation

You’re in the core-agent reference docs. The site root has the marketing pitch; this section is the reference.

Start here

Evaluating core-agent against raw ADK? → Why core-agent is the long-form pitch with a side-by-side capability comparison.

Brand new? → Getting started walks you from go install through core-agent -p "hello" against your first provider, plus the .agents/ project layout.

Running the bundled binary? → Using the CLI splits into Interactive (TUI) — drive the agent yourself from a terminal — and Autonomous (headless) — unattended workers with budgets + crash-resume.

Embedding core-agent in your own Go binary? → Using the library covers the extension points; API is the exhaustive reference.

Tuning prompts, skills, and tool descriptions? → Agent design is the prescriptive section — what patterns work, what failure modes to watch for, how to get the model to use subagents and agentic_* wrappers.

Want an agent to walk you through configuration? → Skills library ships three Claude-Skills bundles covering CLI setup, autonomous setup, and library embedding. Install one and ask your agent for help — same content as the static docs, in workflow form.

Configuring a specific surface? → Reference is the cross-cutting index — providers, config, permissions, MCP, skills, sessions, context management, attach mode.

Looking for a working example? → Examples catalogs every recipe and library quickstart under examples/ in the repo — config-only drop-ins (gke-parallel-triage, plan-first), library quickstarts, autonomous patterns, and testing helpers.

Reference index

CLI

• Using the CLI — interactive vs autonomous landing
  • Interactive (TUI) — quickstart, workflows, slash reference
  • Autonomous (headless) — quickstart, operations, multi-agent GKE scenario

Library

• Using the library — quickstart, guide (narrative tour of extension points), API (exhaustive reference)

Agent design

• Agent design — prompt + skill + tool-description patterns for efficient, well-behaved agents

Skills library

• Skills library — three bundled Claude-Skills (cli-setup, autonomous-setup, library-embedding) so an agent can walk a user through configuration

Examples

• Examples — every config-only recipe (gke-parallel-triage, plan-first) and library quickstart (basic, with-tools, streaming, autonomous patterns, …) with a one-line “use when” summary + link to the repo

Reference

• Configuration — .agents/config.json schema
• Providers — Gemini, Vertex, Anthropic, mock
• Permissions — ask/accept-edits/plan/yolo, patterns, scope
• Built-in tools — the 13-tool model-facing catalog (file / search / shell / network / planning) + lifecycle tools
• MCP servers — declarative third-party tool integration
• Skills — Claude-compatible SKILL.md bundles
• Context management — compaction, checkpoints, agentic tool wrappers
• Sessions and event log — durable storage, audit log, replay, crash-resume
• Attach mode TUI — core-agent-tui remote operator client
• Scion adapter — embedding under Scion’s distributed runtime

Help and community

• Source code → github.com/go-steer/core-agent
• Issues → github.com/go-steer/core-agent/issues — bug reports, feature requests
• Discussions → github.com/go-steer/core-agent/discussions — questions, what-are-you-building threads
• Releases & changelog → latest releases and CHANGELOG.md

What this site doesn’t cover

• Cogo — Claude-Code-style TUI built on top of similar primitives. Different project; see cogo.io docs.
• ADK Go — core-agent wraps Google’s Agent Development Kit. For raw ADK primitives, see the upstream docs.
• Model APIs — for what models can do, see Google AI Studio / Vertex docs and Anthropic’s docs. This site documents how core-agent talks to them, not the model surface itself.