Omd Dream
记忆巩固(同步):经 omd dream_consolidate 把近期事件窗压缩进 L0–L6 记忆层,带节制三闸 + 层级红线 + 健康比自检,只留 derived 洞察不吞业务真相。Trigger:/omd-dream、记忆整理、巩固记忆、consolidate memory、跑一次 dream。
Install
$ agentstack add skill-abysscn-oh-my-dag-omd-dream ✓ scanned · ✓ verified, works with Claude Code, Cursor, and more.
Security review
✓ PassedNo issues found. Passed automated security review. · v0.1.0 How review works →
- ✓ Prompt-injection patterns
- ✓ Secret / credential exfiltration
- ✓ Dangerous shell & filesystem operations
- ✓ Untrusted network calls
- ✓ Known-malicious package signatures
What it can access
- ✓ Network access No
- ✓ Filesystem access No
- ✓ Shell / process execution No
- ✓ Environment & secrets No
- ✓ Dynamic code execution No
From automated source analysis of v0.1.0. “Used” means the capability is present in the source — more access means more to trust, not that it’s unsafe.
Verified badge
Passed review? Show it. Paste this badge into your README, it links to the public security report.
Links to your public security report.[](https://agentstack.voostack.com/security/report/skill-abysscn-oh-my-dag-omd-dream) Reliability & compatibility
Declared compatibility
Compatibility is declared by the source manifest. End-to-end runtime verification is coming, see below.
We're building live execution health for every listing: tool-call success rate, median latency, uptime, and last-checked timestamps, measured, not self-reported. It isn't live yet, so we don't show numbers we can't stand behind.
How agent discovery & health will work →About
/omd-dream — 记忆巩固
对应 omd 的 dream 巩固管线。经 omd MCP 的 dream_consolidate(可能带 mcp__omd__ 前缀;未加载先 ToolSearch "dream_consolidate")把近期事件窗口里的洞察归并进七层记忆(L0–L6:durable fact 带 namespace/confidence)。巩固不是把所有事件都记下来,而是过三道闸、只留少量高价值 derived 洞察。
用法
无参数。适合:长会话收尾、一批任务完成后、owner 说「整理一下记忆」时。dream_consolidate() 同步返回(不走 runId 三段式)但慢(要过模型),调用后耐心等,别超时重发(重复整理 = 重复烧模型)。空窗口不调模型,直接回「无可整理」。
节制三闸(只留 derived,不吞真相)
巩固前每条候选洞察过三关,任一不过即丢:
- UTILITY — 是不是将来用得上的 derived 洞察(偏好/模式/教训/承诺/deadline)?一次性、无复用价值的丢。
- SoT 排除 — 是不是业务真相源(发票/分录/财务数字/文档正文/事实数据)?是就丢——真相住它自己的库,记忆层不做副本,复制 = 日后两份打架。
- 禁忌命名空间 — 落在隐私/敏感类别(个人身份/监控/特殊类别数据)?丢。
留下的应该很少:一轮巩固产出几条 fact 是常态,产出几十条是过度记录的信号。
层级红线(不可绕过)
- L3 业务真相层永不碰 — 任何想写 L3(业务关系/事实数据)的候选直接硬拒 + 记审计,绝不 INSERT/UPDATE/DELETE。
- 写 fact 必经校验闸 — 每条落 durable 层的 fact 都过校验;被拒的 fact 单独记录,不静默塞进记忆。
- 无载体的层走 deferred — 某层暂无存储载体时走 deferred 分支并记审计,绝不伪造「写成功」。
健康比自检
巩固结束看 facts_extracted / events_processed:
- 健康 20% → 过度记录信号,该收紧提炼提示词(在记什么本不该记的?)。
- 转述给 owner 时按层讲「巩固了什么类型的洞察」,有 namespace 冲突或低置信候选提示人工过目。
与既有 skill 的边界
/omd-dream= 把近期事件窗压缩成 durable facts(写入侧,批量、慢、同步)。- 不用于:推理卡住时主动召回既有记忆 →
/omd-recall(读取侧);单条决策/引用手动记台账 →/omd-note。
Source & license
This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.
- Author: AbyssCN
- Source: AbyssCN/oh-my-dag
- License: MIT
Install and usage instructions live in the source repository linked above.
Reviews
No reviews yet, be the first.
Write a review
Versions
- v0.1.0 Imported from the upstream source.