技能 reasoningbank-with-agentdb
📦
reasoningbank-with-agentdb
內容修訂 r1 高風險 ⚙️ 外部命令🌐 網路存取📁 檔案系統存取
使用 AgentDB 建構自適應代理記憶
自我學習代理需要持久記憶來比較決策並改善結果。此技能展示 AgentDB 在軌跡、判斷、蒸餾與檢索方面的模式。
支援: Claude Codex Code(CC)
安裝前請停止並請求確認。
請審查安裝計畫，並在變更檔案前取得使用者明確同意。
使用我的智慧代理安裝
將此請求複製給你的智慧代理，其中包含權威 Skill 頁面和清單。
智能體請求
Review the Skillstore skill "reasoningbank-with-agentdb" from https://skillstore.io/skills/dnyoussef-reasoningbank-with-agentdb.md and its manifest at https://skillstore.io/api/skills/dnyoussef-reasoningbank-with-agentdb/manifest. Verify the artifact. Stop and obtain explicit user consent before installing or changing files.你的智慧代理仍應顯示其計畫，並依安全政策請求必要確認。
Agent 可讀資源
當 AI Agent、爬蟲或腳本需要乾淨脈絡、而不是讀取完整頁面時，請使用這些連結。
測試它
正在使用「reasoningbank-with-agentdb」。 為程式碼審查代理設定自適應記憶。
預期結果:
一份設定計畫，包含先決條件、資料庫位置、記憶領域、初始軌跡欄位與驗證檢查清單。
正在使用「reasoningbank-with-agentdb」。 判斷最佳化軌跡是否成功。
預期結果:
一種使用相似記憶、成功率閾值、信心評分與人工審查觸發條件的裁決方法。
正在使用「reasoningbank-with-agentdb」。 安全地遷移現有記憶資料庫。
預期結果:
一份遷移檢查清單，涵蓋備份、版本釘選、路徑審查、驗證統計與敏感資料控管。
安全審計
高風險v9 • 2026/7/5 開啟版本化報告
Most Ruby or shell backtick alerts are false positives from Markdown fences and TypeScript template literals. Confirmed risks remain for unpinned AgentDB CLI execution, MCP setup, and hidden SQLite memory files that can contain sensitive agent history. No prompt injection or deliberate network exfiltration evidence was found.
1
已掃描檔案
447
分析行數
10
複核項目
0
已忽略的誤報
已確認的安全問題 (10)
高
Unpinned Remote CLI and MCP Setup
The setup instructions run npx agentdb@latest and add an MCP server with claude mcp add. This can execute registry code and persist a local tool integration without version pinning or integrity guidance.
The lines explicitly show latest-version npx execution and MCP registration. The risk is contextual setup execution, not a scanner-only syntax match.
中
SQLite database file
npx agentdb@latest init ./.agentdb/reasoningbank.db --dimension 1536
The command creates a SQLite database intended to hold ReasoningBank memories. Such memories can contain task context, decisions, and sensitive implementation details.
中
SQLite database file
npx agentdb@latest migrate --source .swarm/memory.db
The migration command reads .swarm/memory.db, a legacy SQLite memory store. Agent memory databases can contain sensitive task history.
中
SQLite database file
npx agentdb@latest stats ./.agentdb/reasoningbank.db
The command reads statistics from the ReasoningBank SQLite database. It is likely metadata-only, but it still accesses a sensitive memory store.
中
SQLite database file
dbPath: '.agentdb/reasoningbank.db',
The code configures a persistent SQLite database path for ReasoningBank memory. Persistent agent memory can hold sensitive context if not controlled.
中
SQLite database file
npx agentdb@latest export ./.agentdb/reasoningbank.db ./backup.json
The export command copies the ReasoningBank SQLite database into a backup file. Exported agent memories can disclose sensitive task history if mishandled.
中
SQLite database file
npx agentdb@latest stats ./.agentdb/reasoningbank.db
The stats command reads the SQLite memory database. It is operational metadata access, but the database itself is sensitive.
中
SQLite database file
npx agentdb@latest migrate --source .swarm/memory.db --target .agentdb/reasoningbank.db
The migration command reads .swarm/memory.db and writes .agentdb/reasoningbank.db. Both are SQLite memory stores that may hold sensitive agent history.
顯示全部 10 條已確認問題
中
SQLite database file
npx agentdb@latest stats .agentdb/reasoningbank.db
The stats command reads the AgentDB SQLite memory database. It appears read-only but still targets a sensitive local store.
中
SQLite database file
DEBUG=agentdb:* npx agentdb@latest migrate --source .swarm/memory.db
The DEBUG-enabled migration command reads .swarm/memory.db through external tooling. Verbose logging can increase exposure of sensitive migration details.
能力複核項目 (10)
這些是真實的本機能力，對此技能而言可能是預期行為，因此需要複核，但不會按已確認的惡意行為計分。
高
Ruby/shell backtick execution
```bash
This documented setup block runs npx agentdb@latest and registers an MCP server with claude mcp add. It can execute an unpinned package and modify local tool configuration if followed.
中
Ruby/shell backtick execution
```bash
This block instructs migration and statistics commands through npx agentdb@latest. The commands execute external tooling and touch local memory database files.
中
Ruby/shell backtick execution
```bash
This CLI operations block uses npx agentdb@latest for export, import, and stats. It can execute external tooling and move local memory data.
中
Ruby/shell backtick execution
```bash
This migration block uses npx agentdb@latest to read a legacy database and write an AgentDB target. That is real external command execution over local memory files.
中
Ruby/shell backtick execution
```bash
This troubleshooting block runs shell commands, including DEBUG-enabled npx migration. It is executable command guidance, not merely prose.
中
Ruby/shell backtick execution
```
This learn-more section includes an inline npx agentdb@latest mcp command. It is mostly documentation, but it still presents an external command that could be copied into setup.
中
Hidden file access
npx agentdb@latest init ./.agentdb/reasoningbank.db --dimension 1536
The command creates a hidden .agentdb SQLite database for ReasoningBank data. Hidden local persistence is expected, but it can store sensitive agent memory.
中
Hidden file access
npx agentdb@latest stats ./.agentdb/reasoningbank.db
The stats command reads a hidden .agentdb database path. It appears operational, but it still accesses hidden local state.
中
Hidden file access
npx agentdb@latest export ./.agentdb/reasoningbank.db ./backup.json
The export command reads the hidden .agentdb database and writes backup data. This can copy sensitive agent memories into another local file.
中
Hidden file access
npx agentdb@latest stats ./.agentdb/reasoningbank.db
The stats command accesses the hidden .agentdb database. The operation appears read-only, but it still touches hidden local state.
風險因素
⚙️ 外部命令 (41)
SKILL.md:26-33 SKILL.md:33-37 SKILL.md:37-43 SKILL.md:43-49 SKILL.md:49-95 SKILL.md:95-105 SKILL.md:105-131 SKILL.md:131-137 SKILL.md:137-152 SKILL.md:152-158 SKILL.md:158-189 SKILL.md:189-201 SKILL.md:201-210 SKILL.md:210-211 SKILL.md:211-212 SKILL.md:212-214 SKILL.md:214-220 SKILL.md:220-232 SKILL.md:232-238 SKILL.md:238-247 SKILL.md:247-253 SKILL.md:253-262 SKILL.md:262-263 SKILL.md:263-265 SKILL.md:265-273 SKILL.md:273-294 SKILL.md:294-314 SKILL.md:314-350 SKILL.md:350-356 SKILL.md:356-369 SKILL.md:369-377 SKILL.md:377-387 SKILL.md:387-391 SKILL.md:391-397 SKILL.md:397-404 SKILL.md:404-410 SKILL.md:410-413 SKILL.md:413-420 SKILL.md:420-423 SKILL.md:423-431 SKILL.md:431-439
🌐 網路存取 (2)
📁 檔案系統存取 (4)
審計單位: codex 檢視審計紀錄 →
分享與引用此報告
分享版本化評估報告、中性徽章、嵌入卡片與引用。Skillstore 只報告證據，不判斷該 Skill 是否安全。
複製報告連結
https://skillstore.io/skills/dnyoussef-reasoningbank-with-agentdb/audits/9?utm_source=security_passport&utm_medium=share&utm_campaign=versioned_reportMarkdown 徽章
[](https://skillstore.io/skills/dnyoussef-reasoningbank-with-agentdb?utm_source=security_passport_badge)HTML 徽章
<a href="https://skillstore.io/skills/dnyoussef-reasoningbank-with-agentdb?utm_source=security_passport_badge"><img src="https://skillstore.io/badges/skills/dnyoussef-reasoningbank-with-agentdb/security.svg" alt="Skillstore security assessment" loading="lazy"></a>嵌入卡片
<iframe src="https://skillstore.io/embed/skills/dnyoussef-reasoningbank-with-agentdb.html" title="Skillstore Security Assessment" sandbox="allow-popups allow-popups-to-escape-sandbox" loading="lazy" referrerpolicy="no-referrer" width="420" height="180"></iframe>學術引用 (APA · BibTeX · CFF)
APA 引用
DNYoussef. (2026). reasoningbank-with-agentdb security audit report (audit version 9) [Author version unspecified]. Skillstore. https://skillstore.io/skills/dnyoussef-reasoningbank-with-agentdb/audits/9BibTeX 引用
@techreport{dnyoussef-dnyoussef-reasoningbank-with-agentdb-2026,
author = {DNYoussef},
title = {reasoningbank-with-agentdb security audit report (audit version 9)},
institution = {Skillstore},
year = {2026},
number = {9},
url = {https://skillstore.io/skills/dnyoussef-reasoningbank-with-agentdb/audits/9},
note = {Author version unspecified}
}CITATION.cff
cff-version: 1.2.0
message: "If you use this Skill, cite its author and this versioned security audit report."
title: "reasoningbank-with-agentdb security audit report (audit version 9)"
version: "unspecified"
type: report
authors:
- name: "DNYoussef"
date-released: "2026-07-05"
url: "https://skillstore.io/skills/dnyoussef-reasoningbank-with-agentdb/audits/9"
identifiers:
- type: other
value: "skillstore:dnyoussef-reasoningbank-with-agentdb:audit:9"
description: "Skillstore immutable audit report identifier"
比較變體
2 個可安裝變體每位作者的技能都保留為獨立的可安裝項目。推薦變體依 Skillstore 證據排序。
為何此變體排名第一
Skillstore 評分最高
Skillstore 評分
評分依據 證據置信度: 中 55
架構
85
可維護性
87
內容
71
社群
83
規範符合度
你可以打造什麼
為代理加入學習記憶
建立本機記憶層，用於儲存成功與失敗的決策，以供之後檢索。
評估代理軌跡
在接受自動化裁決之前，將新的任務路徑與相似的既往結果進行比較。
遷移舊版推理資料
透過驗證步驟，將現有 ReasoningBank 記憶移至由 AgentDB 支援的工作流程。
試試這些提示
規劃首次設定
我正在為 [domain] 建構自我學習代理。請指引我建立由 AgentDB 支援的 ReasoningBank 儲存區，並包含先決條件與驗證步驟。
儲存代理經驗
為 [agent task] 設計 ReasoningBank 記憶流程。包含軌跡欄位、成功指標、信心評分與檢索條件。
判斷新軌跡
協助我將這個新軌跡與已儲存的 ReasoningBank 記憶進行比較。建議裁決邏輯、信心閾值與審查防護措施。
治理遷移
建立從舊版 ReasoningBank 記憶遷移至 AgentDB 的計畫。包含驗證、復原、敏感資料處理與版本釘選。
最佳實務
- 釘選 AgentDB 套件版本，並在執行前審查每個 CLI 命令。
- 將記憶資料庫排除在原始碼控制之外，並使用存取控制保護備份。
- 從狹窄領域開始，並在自動化決策前衡量檢索品質。
避免
- 未經審查或版本釘選就執行 latest-version 套件命令。
- 在代理記憶中儲存憑證、私人客戶資料或機密。
- 未經評估或人工審查，就將相似度分數視為最終真相。
常見問題
這項技能協助建構什麼？
它協助使用 ReasoningBank 概念與 AgentDB 儲存，為代理建構自適應記憶工作流程。
它會自動執行 AgentDB 嗎？
不會。它記錄使用者或代理可選擇執行的命令與程式碼模式。
它可以與 Claude Code 搭配使用嗎？
可以。此技能參考 AgentDB MCP integration，以支援 Claude Code 工作流程。
應該儲存哪些資料？
儲存任務上下文、動作、結果、指標與信心值。避免儲存機密與受規範的個人資料。
它本身已可用於生產環境嗎？
不是。若用於生產環境，請加入版本釘選、存取控制、加密、備份與監控。
誰應該使用它？
它適合建構代理記憶、軌跡審查或經驗回放系統的開發者。