Skip to content

Releases: AidaShika/personal-agent-dossier

Release list

v1.0.0

Choose a tag to compare

@AidaShika AidaShika released this 28 Oct 17:16

AI agent platforms keep coming out, and people keep switching. Trying a new platform is easy. Moving the agent is not. personal-agent-dossier 1.0 is a platform-neutral format for that move: the agent's role, anti-jobs, scheduled routines, and long-term memory go into one dossier file (YAML front matter plus Markdown), and an importer on the new platform rebuilds the agent from it, writing down anything that could not be carried over.

In this release

  • The format: SPEC 1.0 (Chinese, with a full English translation in SPEC.en.md). A fingerprint, a short code computed from the content, lets two platforms check they are looking at the same agent without comparing text line by line. Also included: standard test vectors for the fingerprint, a sensitive-data scan, sync semantics, and platform levels L0 to L2. Importers keep accepting the legacy 0.1 and 0.2 spec ids.
  • Four adapters: Grok Bot (the reference implementation), Muse (L1), Hermes (L2, written by Hermes itself), and OpenClaw (L2, written by OpenClaw itself; it found that OpenClaw 2026.6 stores cron jobs in SQLite, not in the legacy JSON file).
  • Two skills, each in Chinese and English: a usage skill and an adapter-writing guide, so an agent on a new platform can seat itself and write its own adapter.
  • Tools: tools/dossier_check.py checks a dossier's fields, fingerprint, and sensitive-data scan. tools/render_notion.py stores dossiers as Notion pages. An importer conformance test is included, and CI runs the vectors, the importers' dry runs, and round-trip smoke tests on every push.

Verified so far

  • Real dossiers round-trip with unchanged fingerprints: the fictional scribe example, and Hatch, a Grok Bot design assistant, moved to Hermes and to OpenClaw.
  • In a cold-start test on 2026-10-29, a freshly installed Hermes with nothing but the public skills received one plain request and completed the Hatch migration on its own, including validation, a round-trip check, and a written report.

Known limits are declared per adapter, in each adapter's own report. Dossiers, this repository, and the reports contain no personal data; exporters refuse to write when the sensitive-data scan hits.

The repository's README and guides have been revised in plain language since this release was tagged. The format itself has not changed.