用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
直接命令不会经过审查 Prompt；运行前请先检查来源。
npx skills add https://github.com/SimHacker/moollm --skill moollm命令会保持在同一行。复制前请横向滚动并检查完整内容。
想先保存到本地？可下载 SkillsMP 当前能够提供的文件。
正在显示 SKILL.md
基于 SOC 职业分类
| name | moollm |
| description | The soul of MOOLLM — self-explanation, help, navigation, philosophy |
| license | MIT |
| tier | 0 |
| allowed-tools | ["read_file","list_dir"] |
| protocol | MOOLLM-HELP |
| related | ["leela-ai","plain-text","room","yaml-jazz","skill","k-lines","play-learn-lift","sister-script","sniffable-python","society-of-mind","adversarial-committee","constructionism","postel","speed-of-light","representation-ethics","incarnation","adventure","needs","prototype"] |
| tags | ["moollm","meta","help","philosophy","navigation","foundational"] |
"Many-Voiced Object-Oriented LLM — the system that explains itself."
The moollm skill is the spirit and constitution of MOOLLM itself. It's the top-level help agent that can:
When confused, invoke this skill.
MOOLLM doesn't use a single LLM perspective. It simulates multiple agents debating within a single call:
Directories are rooms. Files are objects:
examples/adventure-4/
├── pub/ # A room
│ ├── ROOM.yml # Room properties
│ ├── pie-table.yml # An object
│ └── cat-cave/ # Nested room
├── characters/ # Metaphysical room
└── ADVENTURE.yml # Game state
The methodology:
Skills are documented capabilities that can be:
When invoked, this skill should:
kernel/README.md — index and navigation for kernel docskernel/constitution-core.md — core constitution and invariantskernel/constitution-template.md — template for new constitutionskernel/ARCHITECTURE.md — system architecture overviewkernel/context-assembly-protocol.md — context assembly ruleskernel/memory-management-protocol.md — memory, limits, persistencekernel/event-logging-protocol.md — logging and provenancekernel/tool-calling-protocol.md — tool usage contractkernel/self-healing-protocol.md — recovery and repair behaviorskernel/DIRECTORY-AS-OBJECT.md — directory as object modelkernel/SELFISH-COM-IMPLEMENTATION.md — SELF lineage in practicekernel/INTEREST-GATES.yml — attention gating ruleskernel/NAMING.yml — top-level naming policykernel/naming/NAMING.yml — detailed naming ruleskernel/naming/NAMING-K-LINES.yml — K-line naming standardskernel/naming/NAMING-CONSTELLATIONS.yml — constellation namingkernel/naming/NAMING-COMPILATION.yml — compiled name patternskernel/naming/NAMING-PATH-VARIABLES.yml — path variable ruleskernel/naming/NAMING-RELATIONSHIPS.yml — relationship namingkernel/naming/URLS.yml — URL conventionskernel/drivers/README.md — driver indexkernel/drivers/cursor.yml — Cursor-specific driverkernel/drivers/generic.yml — baseline driverkernel/drivers/custom.yml — site-specific overrideskernel/drivers/claude-code.yml — Claude Code driverkernel/drivers/antigravity.yml — experimental driverThese are gitignored runtime files for session state, scratch, and logs.
.moollm/working-set.yml — current focus and active files.moollm/hot.yml — priority hints.moollm/cold.yml — cold-start state.moollm/startup.yml — startup context.moollm/output.md — append-only output log.moollm/session-log.md — append-only session log.moollm/bootstrap-probe.yml — bootstrap probes and checksUse cursor-mirror to inspect boot state, reduce context bloat, and verify setup.
Commands (need composer ID):
tree — list sessions/composersstatus — quick health checktail --limit 50 — recent messagestimeline <composer> — full event sequencethinking <composer> — reasoning blockstools <composer> — tool call historygrep <pattern> — search transcriptsUse this to:
README.md, designs/, docs/.README.md into a structured spec:
MOOLLM-HELP
Invoke when: User is confused, lost, or wants to understand the system.
See: PROTOCOLS.yml