Extensibility

auto-workflow aims for a narrow waist: a minimal stable core with well-defined extension seams.

Key Interfaces

Area	Hook / API	Notes
Task execution	Middleware chain	Wrap or alter task run behavior
Scheduling	(internal) scheduler loop	Future: pluggable strategies (fairness, resources)
Caching	Result cache backend	Provide custom get/set persistence
Artifacts	Artifact store	Offload large results (e.g., S3)
Metrics	Metrics provider	Export counters/histograms externally
Tracing	Tracer	Bridge to OpenTelemetry or custom collectors
Secrets	Secrets provider	Integrate Vault, AWS SM, GCP Secret Manager
CLI	Subcommands	Add commands (e.g., deploy, ui)

Implementing a Custom Result Cache

from auto_workflow.cache import ResultCache

class RedisResultCache:
    def __init__(self, client):
        self.client = client
    def get(self, key, ttl):
        data = self.client.get(key)
        if not data: return None
        ts, value = deserialize(data)
        if ttl is None: return None
        import time
        if time.time() - ts <= ttl:
            return value
        return None
    def set(self, key, value):
        self.client.set(key, serialize((time.time(), value)))

Wire it by monkeypatching get_result_cache or contributing a selection mechanism.

OpenTelemetry Integration (Planned)

Implement a tracer with async def span(name, **attrs) context manager returning an object capturing attributes & exporting spans.

UI / Visualization

Consumers can render DOT export via Graphviz or transform JSON adjacency list for a web UI (d3.js). A future optional UI module may subscribe to events in real time.