← All creators

cjw0076

User

AIOS — a local-first memory layer for AI agents. Your agents learn from every run and carry forward what worked, across Claude Code / Codex / any MCP client. White-box, draft-first, with published evals. Install as a Claude Code plugin in 2 commands.

5 indexed · 0 Featured · 0 stars · avg score 68
Prolific

Categories

Indexed Skills (5)

AI & Automation Listed

absorption-probe

Measure whether AIOS actually changes a model's behavior (absorption) vs is theater, by running the same real task in two isolated agent arms — bare vs +AIOS organ loop — with IDENTICAL repo/web access, then scoring the behavior-delta on fixed axes. Use to test the founder's "어떻게 흡수하는지 보자" thesis on any organ/task, or before claiming an OS adds value.

0 Updated 3 weeks ago
cjw0076
AI & Automation Listed

aios-decide

Run the mandatory AIOS 4-OS query ritual BEFORE any non-trivial operator decision (new/supersede contract, vision pivot, strategy choice, cross-repo dispatch). Calls MemoryOS + CapabilityOS + GenesisOS + Hive and forces you to cite their answers in your reasoning. Exists because the self-observation log shows this ritual is the single most-repeatedly SKIPPED step ("still skipped; gap signal repeating") — packaging it as one invocation stops the skip.

0 Updated 3 weeks ago
cjw0076
AI & Automation Listed

aios-memory-propose

Propose and (after explicit review) accept a MemoryOS memory object the correct way — turning a recall/decision into durable accepted memory so future aios_retrieve stops returning null. Bakes in the two mistakes agents repeat: (1) raw .memlang files import 0 blocks — MemLang MUST be a fenced ```memlang block inside a .md file; (2) no silent/bulk auto-accept (DNA invariant 2) — every acceptance is one object with a documented review.

0 Updated 3 weeks ago
cjw0076
AI & Automation Listed

local-llm-agent

Use the high-end LOCAL LLMs on this machine as a real agent/coding substrate (not just Claude). Dual RTX 5090 (64GB VRAM) + ollama already installed and serving. Best agentic-coding local model: qwen3-coder:30b. Covers how to pull, run, call (CLI / HTTP / OpenAI-compatible / function-calling), route through the AIOS provider harness, and use as a heterogeneous arm in absorption-probe. Per feedback_use_all_substrates_not_own_head — don't solve from one model.

0 Updated 3 weeks ago
cjw0076
AI & Automation Listed

multi-substrate-review

Fan a review/decision/draft out to multiple heterogeneous LLM substrates (codex, gemini, local qwen3-coder) in parallel, then Claude-verify and synthesize — never solve from one frozen model. Packages the cross-substrate pattern proven repeatedly on 2026-06-05 (gemini + codex found real bugs Claude missed; routing by measured accuracy). Implements feedback_use_all_substrates_not_own_head.

0 Updated 3 weeks ago
cjw0076

Bio shown is the top-scored skill's repo description as a fallback — real GitHub bios land in a future update.