🧠 Tiered Recall - 分层回忆系统

分层回忆系统，解决上下文长度限制，保持项目延续性。默认自动加载最近7天记忆，支持手动全量回忆、自定义天数、项目回忆和主题回忆。当前版本采用 slim index，只保留文件名、行号和标题，不存摘要，避免 token 膨胀。

What to check before installing: - Review and fix scripts/check_index.py: it uses a hardcoded absolute path (C:/Windows/System32/UsersAdministrator.openclawworkspace/...) which is likely a mistake and could try to read unexpected system paths. Remove or parameterize that path. - Confirm activation behavior: clawhub.json sets auto activation on session start. If you prefer explicit control, ask the author to disable automatic triggering or require confirmation before running. - Inspect projectPatterns and config (project paths and keyFiles). The loader will attempt to read files under those paths — ensure no sensitive files (credentials, private keys, tokens, secrets) are present in the workspace directories matched by these patterns. - Check .tiered-recall output files (index.json, projects.json, state.json) to ensure they don't accidentally include full text summaries you don't want stored; scripts claim a slim index but check_index.py references a summary field ('s') — verify the actual index content. - Run the scripts in an isolated test workspace first (copy a small set of safe .md files) to observe behavior, outputs, and any unexpected file reads/writes. - If you need stronger guarantees, request the maintainer to: (1) remove/test-only hardcoded paths, (2) document and limit auto-activation, and (3) add explicit checks to avoid reading files outside the intended workspace path. Confidence: medium — the skill appears coherent with its stated purpose but the hardcoded path, manifest activation mismatch, and minor implementation inconsistencies warrant caution and pre-install review.

Type: OpenClaw Skill Name: tiered-recall Version: 1.2.2 The 'tiered-recall' skill is a memory management system designed to help AI agents maintain context across sessions by indexing and retrieving local Markdown logs. Analysis of the Python scripts (build-index.py, load.py) and configuration files shows no evidence of data exfiltration, malicious execution, or unauthorized access. The skill operates entirely within the local workspace as defined in clawhub.json. A hardcoded Windows path in scripts/check_index.py appears to be a developer debugging artifact and does not pose a security risk.