Harneloop v0.0.1 - Alpha

Harneloop v0.0.1 - Alpha

The first alpha release of Harneloop: an agent-first framework for building and improving task-specific AI agent harnesses through evidence-backed trial and error.

Included

• Portable harness units with agent-readable control files and flexible workspaces.
• Agent-first and guided human onboarding.
• Target briefs, environment contracts, and agent-authored attempt plans.
• Artifact-aware run records with immutable finished runs.
• Candidate harness patches developed outside the active harness.
• Parallel candidate batches with impact classification and risk-scaled validation.
• Evidence validation when evidence is attached and again at promotion.
• Restorable promoted versions, rollback, thin packaging, and export adapters.
• Wait, stop, and resume lifecycle states.
• Cross-platform file locking and atomic control-file writes.
• Interactive Rich CLI plus explicit commands for agents and automation.
• GitHub-rendered framework process documentation.
• Permissive Apache-2.0 open-source licensing.

Current Boundaries

• This is an early alpha; commands and file formats may change.
• Installation is currently from the GitHub repository through uv tool.
• The default portable package profile is thin.
• Some global preferences are durable guidance for agents but are not yet enforced by every engine command.
• Harneloop records environment mappings; the operating agent must inspect and map the real tools, endpoints, commands, and artifact paths.

Install

uv tool install git+https://github.com/Ker102/Harneloop.git
harneloop doctor
harneloop setup

See the README for agent-first setup and the complete harness-unit lifecycle.