SkillsMiddleware

Example:

from deepagents.backends.filesystem import FilesystemBackend

backend = FilesystemBackend(root_dir="/path/to/skills")
middleware = SkillsMiddleware(
    backend=backend,
    sources=[
        "/path/to/skills/user/",
        "/path/to/skills/project/",
        # Pass a (path, label) tuple to disambiguate sources whose
        # leaf directories would otherwise collide
        ("/home/me/.claude/skills", "User Claude"),
        ("/repo/.claude/skills", "Project Claude"),
    ],
)

Skill tools

A skill can list the tools its instructions use, separated by spaces, under metadata.include_tools in its SKILL.md frontmatter:

metadata:
  include_tools: create_customer_request list_customer_requests

Pass those tools as tools, either as a list or as a SkillToolResolver that looks them up by name. The model sees a skill tool only after it uses read_file on a skill that lists it, and only while that read stays in context. Until then, calling the tool fails as an unknown tool.

include_tools can also list a tool passed to the agent rather than to this middleware. If that tool is deferred (extras={"defer_loading": True}), reading the skill discloses it automatically.

Placement

create_deep_agent places this middleware for you. When composing create_agent by hand, include FilesystemMiddleware, whose read_file the model uses to read skills. Put this middleware after summarization and any model fallback or routing middleware, so it sees the compacted conversation and the model actually called, and before prompt caching:

create_agent(
    model,
    tools=[...],
    middleware=[
        FilesystemMiddleware(backend=backend),
        ...,
        SummarizationMiddleware(...),
        ModelFallbackMiddleware(...),
        SkillsMiddleware(backend=backend, sources=["/skills/"], tools=[...]),
        AnthropicPromptCachingMiddleware(),
    ],
)

See constructor for the full argument list.