LoopX Capabilities Explained: How to Use Issue‑Fix, Content‑Ops, and ML‑Experiment

TLDR: LoopX provides three core capabilities—issue‑fix, content‑ops, and ml‑experiment—each exposed as modular workflows under loopx/capabilities/ and invoked via the --capability‑route CLI flag or programmatically through packet builders and runners.

The huangruiteng/loopx repository delivers a modular capability system that encapsulates complex automation into reusable components. These LoopX capabilities enable developers to resolve GitHub issues, manipulate web content surfaces, and execute machine learning experiments through a unified interface. Each capability registers itself in a central catalog and exposes both CLI entry points and Python APIs for flexible integration.

What Are LoopX Capabilities?

LoopX capabilities are self‑contained workflow modules stored under loopx/capabilities/. According to the source code, each capability registers a catalog entry in loopx/capabilities/catalog.py containing a unique ID, description, and path to documentation. The system uses a packet‑based architecture where capabilities exchange JSON‑serializable data, enabling composition and replay through the state projection system (loopx/state_projection.py).

Overview of the Three Core Capabilities

Issue‑Fix Capability

The issue‑fix capability automates the full lifecycle of public GitHub issues—from candidate discovery and feasibility checking to reviewer requests and PR creation. It resides in loopx/capabilities/issue_fix/ and provides packet builders such as build_issue_fix_feasibility_packet and validators like validate_issue_fix_feasibility_packet. Users invoke this capability via --capability-route issue-fix or call specific workflow templates directly.

Content‑Ops Capability

The content‑ops capability supplies utilities for web content extraction and surface manipulation. Located in loopx/capabilities/content_ops/, it includes the surface module (loopx/capabilities/content_ops/surface.py) and connectors like loopx/capabilities/value_connectors/social_profile.py. This capability handles social‑browser extraction and surface rendering, activated through --capability-route content-ops.

ML‑Experiment Capability

The ml‑experiment capability provides a lightweight framework for reproducible machine learning workflows. Implemented in loopx/capabilities/ml_experiment/, it exposes helper functions such as run_experiment and log_metrics through loopx/capabilities/ml_experiment/runner.py. Developers activate it with --capability-route ml-experiment to execute training, evaluation, and automated reporting tied to LoopX's state‑projection system.

How to Invoke LoopX Capabilities

LoopX supports two primary invocation patterns: CLI routing for command‑line usage and programmatic access for Python applications.

CLI Routing:

The top‑level loopx CLI parses the --capability-route flag in loopx/cli_commands/ready_score.py and loopx/project_map.py, dispatching commands to the appropriate module. For example:

• loopx --capability-route issue-fix ...
• loopx --capability-route content-ops ...
• loopx --capability-route ml-experiment ...

Programmatic Access:

Each capability exposes packet builders and runners that can be imported directly. These functions create validated JSON packets that capabilities exchange via the state projection system.

Practical Code Examples

Issue‑Fix: Creating a Feasibility Packet

The following example demonstrates building and validating a feasibility packet for GitHub issue #42:

from loopx.capabilities.issue_fix.feasibility import (
    build_issue_fix_feasibility_packet,
    validate_issue_fix_feasibility_packet,
)

# Build a packet for issue #42 in the current repository

packet = build_issue_fix_feasibility_packet(
    issue_ref="#42",
    repo_owner="myorg",
    repo_name="myrepo",
)

# Validate the packet before proceeding

validation = validate_issue_fix_feasibility_packet(packet)
assert validation["is_valid"], validation["errors"]

See the full flow in tests/capabilities/test_issue_fix_feasibility.py.

Content‑Ops: Rendering a Social Profile Surface

Extract and render a LinkedIn profile surface using the social profile connector:

from loopx.capabilities.content_ops.social_browser_x import fetch_profile_surface

# Pull a public LinkedIn profile (the connector knows the source schema)

surface = fetch_profile_surface(username="alice")

# The surface can be rendered as markdown or fed into another capability

print(surface.render_as_markdown())

The connector implementation lives in loopx/capabilities/value_connectors/social_profile.py.

ML‑Experiment: Running a Training Loop

Execute a reproducible experiment using the ML runner:

from loopx.capabilities.ml_experiment.runner import run_experiment

# Define a simple experiment configuration (data, model, metrics)

config = {
    "data_path": "data/train.csv",
    "model": "sklearn.linear_model.LogisticRegression",
    "metrics": ["accuracy", "f1"]
}

# Execute the experiment; results are automatically logged into LoopX state

results = run_experiment(config)
print("Experiment finished – metrics:", results["metrics"])

The experiment driver is located in loopx/capabilities/ml_experiment/runner.py.

Key Files and Architecture

Understanding the file structure helps navigate the capability system:

• loopx/capabilities/catalog.py – Central registry containing capability IDs, descriptions, and documentation paths.
• loopx/capabilities/issue_fix/ – Full implementation including packet builders, CLI templates, and feasibility validators.
• loopx/capabilities/content_ops/ – Modules for content extraction, surface rendering, and social‑browser integration.
• loopx/capabilities/ml_experiment/ – Framework for defining, running, and logging ML experiments.
• loopx/cli_commands/ready_score.py – Parses --capability-route and dispatches to appropriate command packs.
• loopx/state_projection.py – Records all capability actions for replay and auditing.
• docs/capabilities/issue-fix/README.md – User‑facing guide for issue automation.
• docs/capabilities/content-ops/README.md – Documentation for content operations.
• docs/capabilities/ml-experiment/README.md – Overview of machine learning experiments.

Summary

• LoopX capabilities are modular workflows registered in loopx/capabilities/catalog.py.
• Issue‑fix automates GitHub issue resolution via packet builders like build_issue_fix_feasibility_packet.
• Content‑ops handles web extraction and surface manipulation through fetch_profile_surface and related connectors.
• ML‑experiment enables reproducible training via run_experiment in loopx/capabilities/ml_experiment/runner.py.
• Invoke capabilities via --capability-route CLI flag or import Python modules directly.
• All actions are recorded in loopx/state_projection.py for auditability.

Frequently Asked Questions

How do I register a new capability in LoopX?

Register a new capability by adding a catalog entry to loopx/capabilities/catalog.py with a unique ID, description, and documentation path. The entry must point to a valid package under loopx/capabilities/ that follows the packet‑based architecture.

Can I combine multiple capabilities in a single workflow?

Yes. Because capabilities exchange JSON‑serializable packets through the state projection system, you can chain them by passing output packets from one capability (e.g., content_ops surface data) as input to another (e.g., ml_experiment for analysis).

Where are capability results stored for auditing?

All capability actions are automatically recorded in loopx/state_projection.py. This state‑projection system enables full replay, branching, and auditing of any workflow step without manual logging.

What is the difference between packet builders and direct CLI invocation?

Packet builders like build_issue_fix_feasibility_packet are Python functions that construct validated data packets for programmatic use, while CLI invocation via --capability-route triggers the full command‑pack logic defined in loopx/cli_commands/ready_score.py. Use packet builders for custom scripts and CLI routes for standard workflows.

{
  "mcpServers": {
    "instagit": {
      "command": "npx",
      "args": ["-y", "instagit@latest"]
    }
  }
}