v3.3.0 — privacy-safe corpus persistence

Highlights

• applies best-effort secret and PII redaction in memory before the first raw corpus write, including metadata fields
• adds --no-raw / no_raw modes for scans that should create no raw corpus artifact
• expands coverage for private keys, JWTs, AWS, Google, GitHub, Slack, provider, bearer, bot, URL, email, phone, and assignment-shaped sensitive values
• replaces 100-character prefix deduplication with full normalized-content SHA-256, preserving entries with shared openings and distinct tails
• renames the heuristic report to corpus readiness; old quality CLI, MCP, and API names remain compatibility shims
• sanitizes RecallNest stdout before persistence and avoids echoing failed export content
• upgrades the MCP SDK and patched transitive dependencies to a zero-vulnerability npm audit
• adds deterministic regression tests and CI for both CLI and MCP builds

Security boundary

Redaction is best-effort, not a complete secret scanner. Raw and refined corpus still contain private conversation text and must be inspected before sharing or uploading.

Verification

• 5 Bun tests, 41 assertions
• CLI and MCP production builds
• npm audit: 0 vulnerabilities
• GitHub Actions CI on main

v3.1.0

v3.1.0

• Borrowed evidence-grading, capability-boundary, and invocation-rules layers (Stage 4 Soul Forging).
• Docs: ecosystem table cleanup, Do NOT use guidance, measurable_outcome field.
• Aligned package manifest with the published release.

v3.0 Mentor Mode 方法论重构

What's New

借鉴 nuwa-skill 的方法论，聚焦分析角度提升名人克隆准确度。

6 个分析角度（Stage 1 + Stage 4）

角度	解决什么问题
批评者视角	只有自我视角，缺盲区
言行对比	分不清场面话和真信念
三重验证	把随口一说当核心思想
矛盾处理	强行统一导致人格扁平
负空间	不知道边界，遇到就瞎编
争论模式	只有抛光后的思想，缺真实思维过程

流程改动

• Stage 1: 单一 Deep Research prompt → 6 个并行子 agent（按角度分工采集）
• Stage 1.5: 新增调研质量审查检查点
• Stage 2: Mentor Mode 精简为转化+补充
• Stage 4: System Prompt 模板新增 Internal Tensions / Negative Space / Argument Style
• Stage 5: 新增 Voice Test + Negative Space Test + Argument Test + 结构化 Pass Criteria

不变的

• Self Mode 完全不受影响
• src/ 代码无改动
• Stage 3、6 无改动

Version note: v3.0.0 is a repository-level product milestone. In the tagged source, package.json reports 2.0.0; the tag does not claim that a package with the same version was published from that manifest.

v2.0.0

Full Changelog: v1.0.0...v2.0.0

v1.0.0

Digital Clone Skill v1.0.0

AI 数字分身技能，基于写作人格画像驱动。

核心功能

• 从剧评、随笔等文本提取写作人格
• 风格校准与一致性保持
• 与 Content Alchemy 生态集成