APPFL Agent Skill

This repository contains a reusable agent skill for helping users build, configure, run, and customize APPFL applications.

Structure

.
├── SKILL.md
├── agents/
│   └── openai.yaml
└── references/
    ├── configuration.md
    ├── custom-aggregator.md
    ├── custom-trainer.md
    └── metrics-logging.md

• SKILL.md: main skill entry point, trigger description, workflow, and reference routing.
• agents/openai.yaml: UI metadata and default prompt for skill-aware clients.
• references/: focused APPFL guidance loaded as needed for configuration, trainers, aggregators, metrics, metadata, and logging.

Usage

In a skill-aware Codex environment, invoke it explicitly:

Use $appfl to help me configure an APPFL gRPC application.

Example prompts:

Use $appfl to write a custom APPFL trainer that returns train_loss metadata.
Use $appfl to set up APPFL with Globus Compute endpoints.
Use $appfl to define a custom APPFL aggregator for SyncScheduler.

For other coding agents, point the agent at SKILL.md and ask it to use the relevant files in references/.

Maintenance

Keep SKILL.md concise and put detailed examples in references/. After edits, validate the skill with:

python3 /home/abhijit/.codex/skills/.system/skill-creator/scripts/quick_validate.py .