v0.3.0

Added

• A first local discovery prototype for interrupted-task memory evaluation.
• A small memoryvault/ package with local run storage, candidate memory extraction, resume packet generation, and evaluation logging.
• Built-in sample scenarios for coding and document work.
• A JSON intake path for imported interrupted-task traces.
• Example imported traces modeled after coding and long-memory work.
• A synthetic tool-use trace for domain-agnostic interrupted-task testing.
• A Hugging Face benchmark lead registry for later external evaluation.
• A durable-field suggestion step that highlights repeated misses.
• Automated tests covering the local discovery loop.
• A strategy note describing the tool-first, zero-domain-knowledge development phases.
• A Memory Wind Tunnel that removes memory fields and measures the damage.
• A repo-local quality gate with a pre-commit hook, Python linting, Markdown linting, and a 90% coverage threshold.

Changed

• Replaced the placeholder root entry point with a working CLI.
• Promoted assumptions into the resume packet after repeated misses across scenarios.
• Reframed the project from a domain-specific product direction to a tool-first memory-learning direction.
• Updated project docs and .codex notes to reflect the new prototype state.