meta-plan

What problem does it solve?

Meta-plan solves the fragmentation and manual coordination involved in creating, testing, calibrating, and iterating Agents and Skills by providing a single, disk-backed planning and execution controller that supports resumable runs and artifact-driven state.

Core Features & Use Cases

• Structured Plan Generation: Creates timestamped plan markdown files that act as the single source of truth for a create/test/calibrate/iterate workflow.
• Orchestration of Atomic Skills: Drives and invokes meta-* Skills (meta-ideal-state, meta-prompt-engineer, meta-testcase-gen, meta-rubric-gen, meta-eval-judge, meta-debug, meta-iterate) to complete individual steps.
• Resumable Execution & Artifact Management: Recovers context via scripts/context_tool.py, marks step completion based on files in tmp/, enforces backups to bak/, and produces evaluation reports and status.json for synchronization.
• Use Case: Scaffold a new agent, auto-generate testcases and rubrics, run baseline executions across platforms, evaluate with meta-eval-judge, and iterate on prompts based on a consolidated report.

Quick Start

Use meta-plan to create an agent named 'lyrics-bot' and execute a baseline test run to generate a full evaluation report.