Skip to content

CHANGELOG.md

Latest commit

 

History

History
87 lines (52 loc) · 3.28 KB

CHANGELOG.md

File metadata and controls

87 lines (52 loc) · 3.28 KB
Raw

Changelog

2026-05-08

Added

  • Added docs/ORIGIN.zh-CN.md to explain the origin, purpose, usage, and key design turning points of 刻耳柏洛斯系统.
  • Linked the origin document from README.

2026-05-05

Added

  • Reworked README positioning to describe Cerberus as a portable agent skill/protocol, not only a Codex skill.
  • Added usage guidance for both Codex-compatible skill runtimes and other agent frameworks.

Changed

  • Clarified that cerberus-system/ is a Codex-compatible package format, while the core protocol is framework-neutral.

2026-05-05

Added

  • Added language-following policy: Chinese users get Chinese interaction and the product name 刻耳柏洛斯系统; English users get Cerberus System.
  • Added Chinese invocation examples and bilingual UI metadata.

Changed

  • Adjusted product naming guidance to prefer transliteration in Chinese contexts instead of the awkward literal "three-head council" phrasing.

2026-05-05

Added

  • Added model and agent compatibility guidance so Cerberus can run across different LLMs and agent frameworks.
  • Added compact, standard, and deep modes based on model capability and context.
  • Added clearer trigger rules to avoid confusing Cerberus with ordinary pros-and-cons lists.

Changed

  • Simplified skill invocation and OpenAI UI metadata.
  • Clarified that Cerberus does not select a model; it adapts to the active host model.

2026-05-05

Added

  • Added "living parts" guidance so Alpha, Beta, and Gamma speak from protective impulses, fears, desires, sensitivities, and voice texture instead of sounding like mechanical analysis columns.
  • Added living_signature to persona card templates.

Changed

  • Updated Round 1 to include what each head protects, fears, and wants before voting.
  • Updated council guidance to preserve human conflict before translating it into action.

2026-05-05

Added

  • Added mandatory visible cross-questioning requirements for Cerberus councils.
  • Added a council validity checklist to prevent LLMs from compressing the process into a generic summary.

Changed

  • Clarified that Round 2 must run even when the three heads initially agree.
  • Clarified that a valid Cerberus output must show interaction: directed questions, answers, self-blindspots, revised stances, and synthesis.

2026-05-05

Added

  • Added persistent Cerberus Profile support so users do not need to configure the council from scratch every time.
  • Added productization.md with product experience guidance, returning-user flow, profile storage, quick/standard/deep modes, and update discipline.
  • Added identity-neutral Gamma guidance. Gamma is now explicitly user-defined and must not assume gender, family structure, parenthood, marriage, religion, culture, or life path.
  • Added reusable cerberus_profile template.

Changed

  • Updated onboarding flow to ask what the user wants Cerberus to protect from being traded away too casually.
  • Updated adaptive interviewing Gamma choices to avoid family/gender assumptions.
  • Updated README to highlight profile reuse and identity-neutral setup.

Notes

  • Existing Cerberus users should keep their current cards and add a profile file around them.
  • New users can start in quick mode, standard mode, or deep mode depending on available time and energy.