Menú
Building agents with BaseAgent / GenericAgent, including skill registration, activation modes (llm vs manual), and system-prompt composition. Use when editing src/echo/agents/ or building a new agent subclass.
Instalar con Codex o Claude Copia este prompt, pégalo en Codex, Claude u otro asistente, y deja que revise la página de la skill y la instale por ti.
Basado en la clasificación ocupacional SOC
Router for any non-trivial echo-sdk task. Maps user intent to the right module skill and diagram. Invoke FIRST for any change to src/echo/ before reaching for a specific module skill.
Recipe for adding a new provider (LLM, transcriber, eval, prompt). Use whenever adding Cohere, Mistral, Deepgram, Phoenix, Promptlayer, etc. — anything that plugs into an existing get_*() factory.
End-to-end recipe for defining a Skill bundle and wiring it into a GenericAgent with the right activation mode and tool registry. Use when extending agent capability via swappable skills rather than baking tools in.
Audio transcription provider abstraction (Gemini, EkaCare). Use when transcribing audio, adding a new transcriber, or shaping AudioInput.
Postgres support — asyncpg client, schema-aware binder, PgQueryTool surface. Use when querying postgres from a tool, modifying the binder, or adding a new DB engine.
Evaluation provider abstraction for dataset experiments (Langfuse today). Use when running an experiment, adding a new eval backend, or wiring evals into CI.
| name | echo-sdk-agents |
| description | Building agents with BaseAgent / GenericAgent, including skill registration, activation modes (llm vs manual), and system-prompt composition. Use when editing src/echo/agents/ or building a new agent subclass. |
BaseAgent (abstract) in src/echo/agents/base.py — owns tool registry, skill registry, LLM, system-prompt composition.GenericAgent in generic_agent.py — the concrete 90% path; YAML-config-driven.Skill in skill.py — bundle of (name, description, instructions, tool_names). Subclass only for on_activate / on_deactivate hooks.AgentConfig / PersonaConfig / TaskConfig in config.py + load_agent_config() for YAML.AgentResult in schemas.py.BaseAgent only when GenericAgent doesn't fit. Most use cases just instantiate GenericAgent(agent_config=..., llm_config=..., tools=..., skills=...).load_skill / unload_skill are silently dropped with a warning.Skill.tool_names must resolve in the agent's tool registry at construction — unresolved skills are dropped with a warning.skill_activation:
"manual" — host calls await agent.activate_skill(name, context)."llm" (default) — load_skill / unload_skill meta-tools are auto-injected.base_tool_names: explicit empty list [] means "nothing default-visible — everything is skill-gated." None means "all non-meta tools visible by default."base ∪ ⋃(active skill.tool_names) ∪ meta. See _build_active_tools in base.py.on_activate / on_deactivate receive Optional[ConversationContext] — None during setup, live context during a turn.When skills are registered, the prompt becomes:
<base_user_prompt>...role/goal/task...</base_user_prompt>
<skill_mechanism>...</skill_mechanism>
<available_skills>...</available_skills> (llm mode only)
<active_skills>...</active_skills>
<active_skill name="x">instructions</active_skill>
The mechanism template lives in src/echo/prompts/templates/ (loaded by load_template("skill_mechanism")).
async def run(context, out_msg_id, **kwargs) -> AgentResult — non-streaming.async def run_stream(context, out_msg_id, **kwargs) -> AsyncGenerator[StreamEvent, None] — streaming; terminal DONE event holds full response + updated context.Both must catch exceptions and return/yield an error result (AgentResult(error=...) or StreamEvent(type=ERROR)) — never let the agentic loop bubble unhandled.
self.tools mid-conversation → don't; tool list is computed per turn from the registry.Skill.on_activate to fetch tenant-scoped tokens at activation time.activate_skill before agent setup completes without passing context=None → pass None explicitly during setup.llm mode — for very small contexts, prefer manual activation..claude/diagrams/agent-skill-lifecycle.md[[echo-sdk-llm]], [[echo-sdk-tools]], [[echo-sdk-models]], [[echo-sdk-adding-a-skill-to-an-agent]]