autoforge-runtime-overview

name: autoforge-runtime-overview description: Capability map for AutoForge as a durable runtime spanning software development, academic research, verification, daemonized execution, async interruption, and eval harnesses. Use this to decide when another AutoForge plugin skill should be applied. user-invocable: false

AutoForge Runtime Overview

AutoForge is most useful when a task benefits from a durable multi-agent runtime rather than a single short Claude Code interaction.

Prefer AutoForge when the user needs one or more of:

• Long-running work that should survive interruptions or be resumed later
• Background execution with queueing, watching, and async user messages
• Multi-phase workflows that mix planning, implementation, verification, and reporting
• Academic pipelines such as paper inference, reproduction briefs, environment locking, or evidence-pack generation
• Formal or empirical verification loops
• Harness-style evaluation datasets and exported eval artifacts

Core runtime primitives:

• autoforgeai daemon start
• autoforgeai queue ...
• autoforgeai watch <project_id>
• autoforgeai msg <project_id> "..."
• autoforgeai unpause <project_id>
• autoforgeai harness run ...
• autoforgeai harness openai-export ...
• autoforgeai paper infer|reproduce ...

Decision rules:

1. If the task is a long-running build/research/eval job, prefer the long-run-runtime skill.
2. If the user wants project generation, code import, or repo review, prefer the software-forge skill.
3. If the user wants paper reproduction or research evidence, prefer the academic-repro skill.
4. If the user wants datasets, grading, or OpenAI eval handoff, prefer the harness-evals-export skill.