360 Secure SkillHub
360 安全技能中心

session-atlas

该 Skill 通过本地 Python 脚本读取指定 Claude JSONL 会话与输出目录,生成含提示、工具预览和推理哈希的私有索引,并提供结构或完整工具链校验;还可依据方法论与模板编写区分观察、解释、决策和证据缺口的导航文档。索引含敏感内容,需私密存放。

antreasantoniou
antreasantoniou
热度 7.64
最近更新 2026-09-12 02:55:25
分类:知识依赖:代码执行 / 读写文件应用场景:笔记整理 / 数据处理 / 工作流 / 研究

安装技能

复制下面的话术发给你的 Agent,它会通过 secure-skills 检索并安装该技能。

执行命令安装此 skill: npx secure-skills-cli add session-atlas-agent-skill
Metadata
name
session-atlas
description
Build a source-indexed navigation layer over a private Claude JSONL session, preserving evidence locations and uncertainty without reproducing private reasoning.

Session Atlas

Use scripts/index_claude_session.py with explicit --source and --out-dir paths. Output includes private prompt/assistant text and tool previews: it is NOT anonymized or publication-safe. Choose a private directory outside public repositories. Raw reasoning blocks are represented by hashes and counts only. Validate with scripts/validate_atlas.py --index-dir DIR --source SOURCE. Add --require-complete-tools when a complete tool trail is required. Without --source, validation is structural only. A valid index does not establish factual or semantic accuracy. For a human-authored map, use references/methodology.md and templates/atlas.md. Keep source observations, interpretations, decisions, outstanding work and evidence gaps distinct. Link claims to snapshot hashes and line spans. Preserve the difference between requested, prepared, executed and verified. Do not publish source transcripts, generated indexes or personal paths without separate review and authorization. No external service or private companion skill is required.