agentic-runtime

name	agentic-runtime
description	Use this skill when working with the Agentic Elixir library. Consult for agent loop configuration, profiles, tools, callbacks, strategies, and testing patterns.

Overview

Agentic is a composable AI agent runtime for Elixir (~> 1.19). It provides a stage pipeline architecture where middleware-style stages wrap each other to process agent turns.

Quick Reference

Entry Points

• Agentic.run(prompt:, workspace:, callbacks:) — primary entry point
• Agentic.resume(session_id:, workspace:, callbacks:) — resume a session
• Agentic.new_workspace(path) — scaffold a workspace

Critical Convention: String Keys

All tool schemas, messages, content blocks, and LLM responses use string keys.

Correct Tool Names

read_file, write_file, edit_file, list_files, bash, delegate_task, skill_list, skill_read, skill_search, skill_info, skill_install, skill_remove, skill_analyze, memory_query, memory_write, memory_note, memory_recall, search_tools, use_tool, get_tool_schema, activate_tool, deactivate_tool

Callbacks

Only :llm_chat is required. All others are optional.

Profiles

:agentic (default), :agentic_planned, :turn_by_turn, :conversational, :claude_code, :opencode, :codex, :acp

Phase Machine

Phases: :init, :plan, :execute, :review, :verify, :done. Transitions go through Agentic.Loop.Phase.transition/2 — never mutate directly.

References

See the reference files for detailed information:

• tools.md — Complete tool schemas
• ../usage-rules/loop.md — Pipeline stages and profiles
• ../usage-rules/callbacks.md — All callback signatures
• ../usage-rules/protocols.md — AgentProtocol behaviour
• ../usage-rules/strategies.md — Strategy behaviour
• ../usage-rules/testing.md — TestHelpers