An actual 0.10.0 synthetic-data result: adapter creation, a context pack, and verification without saved-state changes.
Last reviewed:
About this guide
This guide was prepared with AI assistance using the project’s public documentation. It does not claim testing on every device or independent human review. Project publisher: AmirSina Zamanian.
Sources are listed at the end. Use the project repository to report an error or suggest a correction. Report an issue
⌕
Search runs on this device.
On this page
What was the result?
On September 12, 2026, the public personal-agentic-os 0.10.0 wizard ran on macOS with Node.js v24.19.0 in dashboard-only mode using programmatic answers. It returned ready, created AGENTS.md and CLAUDE.md, resolved two same-path candidates to one project, and generated one context pack. Verify returned healthy without changing state.json. The interactive CLI was also exercised in a separate fixture by answering its prompts one at a time; final verification was healthy.
All inputs were synthetic: a demo project in a temporary directory, empty usage sources, manual sync, and disabled telemetry. The plan was declined first and absence of config.json checked; it was then approved. A sample decision with synthetic-lab provenance was added to fixture memory and synced. The script asserts the decision text, provenance, and Verified: true in the output. That verification label belongs to the test fixture, not to a user-project fact.
Download and read the script below. In an empty experiment directory, run npm install --ignore-scripts --no-audit --no-fund personal-agentic-os@0.10.0; fetching the package requires network access. Then run node handoff-lab.mjs ./node_modules/personal-agentic-os. The script creates a fresh temporary fixture directory and prints its location. Supply no personal data. A zero exit code and final JSON mean its assertions passed. The fixture directory remains available for inspection.
No live Codex or Claude Code session, Obsidian interface, real usage source, or other operating system was tested. Interactive setup was exercised only in dashboard-only mode with synthetic inputs. Identical adapter text does not prove that two models understand it equally. Token, cost, time, and traffic improvements were not measured. This is a local installation-and-context-pack experiment, not an AI performance benchmark.