Tiered Recall

分层回忆系统 - 解决上下文长度限制，保持项目延续性。每次新session自动加载核心记忆+最近日志+活跃项目，支持手动深度回忆。索引含10字内摘要，方便区分同名条目。

This skill appears to implement a local 'tiered recall' system and mostly behaves as described, but review and fix two issues before trusting it with real data: 1) scripts/check_index.py contains a hardcoded Windows System32 path (C:/Windows/System32/UsersAdministrator.openclawworkspace/.tiered-recall/index.json) that is inconsistent with the workspace-local model and could read unexpected files — remove or adapt that path to your workspace or delete the script. 2) The documentation mentions update-projects.py but that file is missing; expect runtime errors or incomplete features. Also: the scripts read and print local files (MEMORY.md, recent logs, project key files and previews). Although these scripts don't make network calls themselves, any sensitive data in those files could be exposed by the agent if it forwards content externally. Recommended actions: run the code in a sandbox or review/modify the scripts (remove/check hardcoded paths, add input validation, handle missing keys safely) before installing; test on a non-sensitive workspace; and prefer to keep any secrets out of the memory/ memory/ directory if you enable automated loading. My confidence is medium — the odd hardcoded path and missing file are clear red flags but do not prove malicious intent on their own.

Type: OpenClaw Skill Name: tiered-recall-memory Version: 1.1.0 The 'tiered-recall-memory' skill is a context management utility designed to help AI agents maintain project continuity across sessions by indexing and loading local markdown files (logs and core memory). The Python scripts (build-index.py, load.py) perform standard file I/O operations within the workspace to categorize information into tiers (L0-L3) and do not contain any network calls, obfuscation, or unauthorized data access. The instructions in SKILL.md are consistent with the stated purpose of improving context retrieval and do not exhibit malicious prompt-injection patterns.