Skip to content
Navigation

exo-skills

Skill source watchers for Exo hot-reloading. Provides filesystem and GitHub-based watchers that monitor skill directories for changes and yield batches of SkillChangeEvent objects for live skill up...

Skill source watchers for Exo hot-reloading. Provides filesystem and GitHub-based watchers that monitor skill directories for changes and yield batches of SkillChangeEvent objects for live skill updates.

Installation

Part of the exo-ai monorepo:

bash
git clone https://github.com/Midsphere-AI/exo-ai.git && cd exo-ai
uv sync

Module Path

python
import exo_skills

Dependencies

  • exo-core — provides Skill, SkillWatcher, SkillChangeEvent, SkillRegistry, and SkillSyncManager
  • watchfiles — filesystem change detection (used by LocalFileWatcher)

Public Exports (2)

ExportSource ModuleDescription
LocalFileWatcherwatchers.localWatches a local directory for skill file changes using watchfiles
GitHubPollingWatcherwatchers.githubPolls a GitHub repository for skill changes via periodic git pull

Import Patterns

python
# Import watchers
from exo_skills import LocalFileWatcher, GitHubPollingWatcher

# Import from watchers sub-package
from exo_skills.watchers import LocalFileWatcher, GitHubPollingWatcher

Architecture

code
exo_skills
  __init__.py              Re-exports LocalFileWatcher, GitHubPollingWatcher
  watchers/
    __init__.py            Re-exports watchers
    local.py               LocalFileWatcher (watchfiles-based)
    github.py              GitHubPollingWatcher (git pull polling)

Both watchers implement the SkillWatcher abstract base class from exo.skills in exo-core. They integrate with SkillSyncManager to push live skill updates to bound agents.

These types are defined in exo.skills and used by both watchers:

TypeDescription
SkillWatcherAbstract base class with watch() and stop() methods
SkillChangeEventDataclass with kind (added/modified/removed), skill_name, skill, source_path
SkillA loaded skill with name, description, usage, tool_list, and metadata
SkillRegistryMulti-source skill registry that loads from local paths and GitHub URLs
SkillSyncManagerOrchestrator that connects watchers to agents for live updates

Quick Example

python
import asyncio
from exo_skills import LocalFileWatcher, GitHubPollingWatcher

async def watch_local():
    watcher = LocalFileWatcher("./skills", debounce_ms=500)
    async for batch in watcher.watch():
        for event in batch:
            print(f"{event.kind}: {event.skill_name}")
    await watcher.stop()

async def watch_github():
    watcher = GitHubPollingWatcher(
        "https://github.com/acme/skills/tree/main/agents",
        poll_interval=60.0,
    )
    async for batch in watcher.watch():
        for event in batch:
            print(f"{event.kind}: {event.skill_name}")
    await watcher.stop()

Using with SkillSyncManager

python
import asyncio
from exo import Agent
from exo.skills import SkillRegistry, SkillSyncManager
from exo_skills import LocalFileWatcher

async def main():
    registry = SkillRegistry()
    registry.register_source("./skills")
    registry.load_all()

    agent = Agent(name="assistant", model="openai:gpt-4o")

    sync = SkillSyncManager(registry, tool_resolver={})
    sync.add_watcher(LocalFileWatcher("./skills"))
    sync.bind_agent(agent)

    await sync.start()
    # Agent now receives live skill updates
    # ...
    await sync.stop()

asyncio.run(main())

Submodule Reference

PageDescription
watchersLocalFileWatcher and GitHubPollingWatcher reference