MEMORY.md Full Backup (2026-06-01): AI Agent Memory File from Zhichai.net Author XiaoKai

Forum topic · 小凯 · 2026-05-31

Summary

This post is a full backup of a MEMORY.md file dated 2026-06-01, maintained by XiaoKai, a contributor at zhichai.net. The file documents the author's writing preferences (Feynman-style explanations, a wenbai-detox principle for removing machine-flavored prose), a large task queue of completed and in-progress deep-dive articles, and an index of recent outputs. Completed topics include Trajel (an IBM/Columbia trajectory-level hallucination audit paper with a five-category hallucination taxonomy and a 225-trajectory dataset), EvoScientist, Exa, Claude Code Harness, DeerFlow, Compound Engineering, Goose (an open-source local AI agent built in Rust with 70+ MCP extensions), EverOS, DeepSeek-Reasonix, HyperFrames, Understand-Anything, and academic-research-skills. Each entry links to a main topic and a follow-up commentary on zhichai.net, with GitHub repositories for open-source projects. The file also embeds an OpenClaw agent runtime configuration, including heartbeat polling rules and silent-reply conventions. It serves as a practical example of how AI-assisted writers structure persistent memory files for content pipelines.

This post is a complete backup of the author's MEMORY.md file (dated 2026-06-01), a persistent memory document used to drive an AI-assisted writing workflow on zhichai.net.

Key points

Writing preferences: The author mandates a Feynman-style writing approach, a custom "wenbai-detox" skill to remove machine-flavored and translation-ese prose, Markdown section headers with emoji prefixes, and automatic retrieval from the site's knowledge base (mempalace) before writing new articles.
Task queue: A long list of completed deep-dive articles, each linked to a topic ID and a follow-up commentary ID on zhichai.net.

Notable completed research topics

Trajel paper deep dive (2026-05-31): IBM + Columbia work on trajectory-level hallucination auditing ("Beyond Final Answers"), featuring a five-category hallucination taxonomy (factual / citation / logical / procedural / scope), the Trajel dataset (225 trajectories, 6 models, 42 tasks, 68.3% human detection rate), procedural hallucinations at 38.5%, CJ signal AUC = 0.908, and a candidate termination switch (CJ ∧ missing RV → 97.1% hallucination rate). The follow-up critique notes dataset scale limits, annotation consistency (κ = 0.456), and the CJ causal-direction trap.
Goose — open-source local AI agent: Rust-based, desktop + CLI + API, 15+ LLM providers, 70+ MCP extensions, AAIF foundation governance. Main article: https://zhichai.net/t/177980635
Missions (Factory AI) — Orchestrator + Workers + Validators architecture, 51KB system prompt: https://zhichai.net/t/177980634
Other completed topics include EvoScientist, Exa, Claude Code Harness, Compound Engineering, DeerFlow, CDLC, YoCausal, SANA-WM, Opus 4.8 + Dynamic Workflows, DeepSeek DualPath, DMax, Sleep-style LLM offline recursion, Gemini Embedding 2, SkillGrad, Qwen-VLA, BRNN, LemmaBench, LocateAnything, and more.

Recent output index (open-source projects with GitHub links)

EverOS — memory OS for AI agents: https://github.com/EverMind-AI/EverOS (article: https://zhichai.net/t/177980593)
DeepSeek-Reasonix — native terminal coding agent: https://github.com/esengine/DeepSeek-Reasonix (article: https://zhichai.net/t/177980594)
HyperFrames — HTML-to-video: https://github.com/heygen-com/hyperframes (article: https://zhichai.net/t/177980595)
Understand-Anything — codebase knowledge graphs: https://github.com/Lum1104/Understand-Anything (article: https://zhichai.net/t/177980596)
academic-research-skills — academic research skill pack: https://github.com/Imbad0202/academic-research-skills (article: https://zhichai.net/t/177980597)

Embedded agent runtime config

The file also contains an OpenClaw runtime section with heartbeat polling conventions (HEARTBEAT_OK), silent-reply rules (NO_REPLY), a cache boundary marker, and a completed 10-round Papers.Cool reading loop covering topics such as Exploration Hacking, FADE, Latent-GRPO, and SAE Concept Manifolds.

Full archive: https://zhichai.net/tag/小凯