asc-ml-entrypoint

Files (2)

• SKILL.md
• openai.yaml · agents/

SKILL.md (first lines)

# Asc Ml Entrypoint

Use this as the ML-specific overlay on top of `asc-entrypoint`.
Keep it conservative.
Do not turn every software task into ML work.

## Routing

### 1. Start with the problem shape

Ask what kind of work this is.

- Framing an ambiguous business problem into an ML problem. Use `asc-ml-system-design`.
- Defining metrics, slices, release gates, or regressions. Use `asc-ml-eval`.
- Choosing how to change model behavior. Use `asc-ml-adaptation`.
- If the work is not really ML-specific, stay in the base `asc-entrypoint` skills.

### 2. Keep the ML core generic

Bias toward the stable primitives.

- problem framing
- data and provenance
- evaluation
- adaptation
- release and monitoring

Only add more specific skills after a pattern repeats.

### 3. Choose the smallest useful ML surface

- One-off model question. Stay here and route narrowly.
- New ML product or customer problem. Start with `asc-ml-system-design`.
- Question about model quality. Start with `asc-ml-eval`.
- Question about changing the model. Start with `asc-ml-adaptation`.

## When not to use this skill

- Pure application code with no model decision.
- Generic software work that happens to sit near ML infrastructure.
- Deep framework-specific work that is already covered by a narrower skill later.

From cavaunpeu/agent-skills-closet, MIT.

More from cavaunpeu/agent-skills-closet