agents/

Reference scaffolds for third-party agents that integrate with MisakaNet.

The current files in this directory are examples, not production agents — they exist so a new operator can copy-paste a starting point without reinventing the shape of a MisakaNet-compatible agent.

your_agent.py — minimal lesson-aware agent

A 29-line stub demonstrating the minimal contract that any MisakaNet agent is expected to satisfy:

Method	Purpose
__init__(name)	Take a stable identifier (used as the agent_id in usage/telemetry).
run(task, lesson=None)	Execute one task; pass a prior lesson (or list of lessons) when you want the agent to consult MisakaNet knowledge before acting.
get_lesson(result)	Convert a run() result into a lesson-shaped dict so it can be fed back into a future run() call.

your_agent.py is intentionally not wired to any LLM provider. To turn it into a working agent:

1. Implement the actual task execution inside run() (call a model, call tools, whatever the agent's job is).
2. Replace the synthetic success/failure result with a structured dict so get_lesson() can extract useful failure metadata.
3. Emit lessons in the shape documented under docs/agents/knowledge-structure.md (frontmatter + body).

When this file is NOT the right starting point

• You want a server-side lesson search/MCP integration. Use scripts/mcp_server.py (the MCP server) or search_knowledge.py, not a custom agent.
• You need a real multi-tenant agent runtime. Look at the lessons under lessons/contrib/hermes-* and lessons/core/hermes-* for the production Hermes Agent shape this directory was scaffolded from.

Related lessons

• lessons/contrib/hermes-model-switch-ccswitch.md — Hermes/CC model switching conventions.
• lessons/contrib/misakanet-refactor-v2-review.md — how the package was trimmed down; your_agent.py predates that refactor.