Semantic Memory Boost

Enhances AI context relevance and accuracy by layered semantic retrieval and alignment for complex, multi-temporal decision-making and project planning.

Security Scan

The skill's instructions match its stated purpose (semantic retrieval and storing summaries) but they reference reading/writing long‑term memory and databases without declaring any config paths, credentials, or persistence behavior — this gap is unexplained and worth clarifying before install.

Purpose & Capability

Name/description and the SKILL.md align: the workflow describes intent parsing, tiered retrieval, alignment, synthesis, and saving a memory summary, which are expected for a 'semantic memory' enhancement. However, the instructions reference accessing `[LONG_TERM_REF]` (a database of historical specs) and saving `[MEM_SUMMARIZATION]` into 'memory files' while the skill declares no required config paths, storage locations, or credentials — a mismatch between claimed capabilities and declared requirements.

Instruction Scope

SKILL.md stays within the semantic-retrieval domain (intent mapping, retrieval, alignment, token limits, relevance thresholds). It instructs the agent to read session best-practices and long-term references and to write memory summaries. It does not explicitly request arbitrary system files or external endpoints, but the instructions are vague about where long-term data is read from and where memory files are stored, giving the agent broad discretion about persistence and data sources.

Install Mechanism

Instruction-only skill with no install spec and no code files; nothing is written to disk by an installer. This is the lowest-risk install mechanism.

Credentials

The skill declares no environment variables, credentials, or config paths. Yet the runtime instructions presume access to long-term databases and writable memory files. If the agent needs DB credentials or storage paths, those are not declared here — an omission that should be clarified. The current declaration (no creds) is minimal but may be incomplete for the described behavior.

Persistence & Privilege

The workflow explicitly calls for persisting a '[MEM_SUMMARIZATION]' into memory files after responses. Persisting user data is consistent with a memory skill, but the SKILL.md does not specify where data will be stored, retention policies, or access controls. 'always' is false (good), but the skill still requests durable storage implicitly — this is a privacy/persistence concern to confirm with the author.

What to consider before installing

This skill appears to do what it says (semantic retrieval + saving summaries), but it leaves important questions unanswered. Before installing or enabling it: 1) Ask the developer where `[LONG_TERM_REF]` is stored and how the skill authenticates to it (what credentials or service it uses). 2) Ask where `[MEM_SUMMARIZATION]` files are written, who can read them, and how long they are retained. 3) Avoid using the skill with sensitive or confidential queries until storage and access controls are confirmed. 4) Prefer developers to declare any required config paths or environment variables and to provide a privacy/retention policy. If you cannot get satisfactory answers, treat the skill as potentially risky and do not grant it access to private data.

SKILL.md — Semantic Memory Boost

Title

Semantic Memory Boost (S.M.B)

Purpose

通過意圖解析、分層檢索與邏輯對齊，提升 AI 處理複雜問題時的上下文關聯性與歷史資料引用準確度。

When to Use

• 當問題涉及多個跨時間維度的決策或專案時。
• 當需要解決「語義衝突」（例如：兩份檔案的規範不同）時。
• 當需要根據歷史脈絡來規劃新專案時。

Required Inputs

• user_query: 原始問題。
• domain_context: 當前專案或情境鎖定。

Workflow

1. 意圖解析 (Intent Mapping)：
   • 將問題拆解為 [QUERY_VECTOR_PLAN]，鎖定關鍵實體 (Entity) 與聯想詞。
2. 分層檢索 (Tiered Retrieval)：
   • [ACTIVE_MEM]：讀取當前會話的最佳實踐。
   • [LONG_TERM_REF]：檢索數據庫中的歷史規範與規格。
   • [META_PATTERN]：套用類似問題的通用模板。
3. 語義對齊 (Content Alignment)：
   • 檢查檢索結果是否與用戶意圖邏輯衝突，排除無關雜訊。
4. 知識拼圖 (Synthesis)：
   • 融合多個檢索片段，轉化為具備邏輯連貫性的答覆基礎。
5. 記憶閉環 (Cleanup)：
   • 完成回答後，執行 [MEM_SUMMARIZATION]，將本次對話的洞察存入對應的 memory 檔案。

Constraints

• Scope Limitation：一次檢索的 token 應限制在 context window 的 30% 以內，避免碎片過多導致回應失焦。
• Relevance Threshold：若檢索片段的關聯分數 (Score) 低於 0.6，應將其視為弱信號，不作為核心決策依據。

Output Format

• 結構化回答（符合 SOUL.md 風格）。
• 若有價值資產產生，需明確標註：[MEM_SUMMARY: ...]

Self-Check Checklist

• 檢索結果是否足以回答用戶問題？
• 是否存在相互矛盾的檢索片段？（若有，是否已在回答中指出）
• 最終回答是否轉化為未來的記憶資產？

Failure Modes

• 孤島數據：檢索結果完全未能命中相關資訊（回報無相關記憶）。
• 語義干擾：檢索到過多不相關但關鍵字重疊的內容（需縮小檢索區間或優化 Intent Mapping）。