← 返回 Skills 市场
366
总下载
0
收藏
2
当前安装
2
版本数
在 OpenClaw 中安装
/install subagent-distiller
功能描述
自动增量提取对话中的结构化知识，智能过滤无用信息，动态聚类主题，支持状态追踪和长期价值沉淀。
安全使用建议
This skill appears internally coherent and implements what it claims, but take these precautions before installing/running:
- Backup your memory/topics and related workspace directories (the scripts will move/archive topic files and create/overwrite cards).
- Update the hard-coded paths (/home/aqukin/...) in the scripts to match your environment to avoid accidental edits in the wrong home directory.
- Confirm which agent/model/process will execute the extraction tasks (extraction_tasks.json / sessions_spawn). The system will send raw conversation slices to that processor, so ensure the model/endpoint is authorized and acceptable for processing sensitive data.
- Clarify the repository URL inconsistency in SKILL.md vs package.json and prefer obtaining the code from a trusted source (verify upstream repo/commit history).
- Run in a test environment first (or run incremental_slice and realtime_distill in dry-run / inspect-mode) to observe outputs before adding cron jobs or using bulk_cleanup --exec.
- Note: bulk_cleanup prints what would be deleted and requires --exec to perform moves, but review the printed list carefully before executing.
If you want, I can list the exact files/lines that reference hard-coded paths and suggest safe edits to make the skill use a configurable workspace path.
功能分析
Type: OpenClaw Skill
Name: subagent-distiller
Version: 3.0.1
The skill bundle contains hardcoded absolute paths to a specific user's home directory (`/home/aqukin/`) across all core Python scripts, including `bulk_cleanup.py`, `realtime_distill.py`, and `incremental_slice.py`. While the logic for scanning session logs and extracting structured knowledge appears functional and aligned with the stated purpose, the reliance on fixed, user-specific paths is a significant security risk and suggests the code was not properly generalized for public use. No clear evidence of data exfiltration or intentional malice was found, but the hardcoded environment and broad file manipulation capabilities make it suspicious.
能力评估
Purpose & Capability
Name/description (incremental distillation of conversation memory) aligns with the included scripts: incremental_slice.py reads session jsonl and produces slices, realtime_distill.py prepares extraction tasks and finalizes cards, domain_consolidate.py merges domains, lifecycle_manager.py manages reminders, and bulk_cleanup.py re-evaluates/archives cards. No unrelated credentials, binaries, or services are requested.
Instruction Scope
SKILL.md and scripts instruct the agent/operator to read conversation session files, create slices, generate extraction tasks, and rely on a 'main agent / sessions_spawn' subagent to run the prompts — meaning raw conversation content will be sent to whatever model/process handles those tasks. bulk_cleanup.py can archive (move) many topic files (requires explicit --exec to perform deletes), and crontab instructions schedule automatic runs. Also note a minor inconsistency in SKILL.md's example git clone URL (github.com/yourname/...) vs package.json repository (github.com/openclaw/...), which should be clarified before installing.
Install Mechanism
No install spec or remote downloads; this is instruction-and-script based. No brew/npm/remote archive downloads are performed by the repo itself. Scripts operate on local files only.
Credentials
The skill requests no environment variables or credentials (proportional). However all scripts use hard-coded absolute paths under /home/aqukin/.openclaw/workspace and related dirs — you must adjust those to your environment. The scripts read session logs (sensitive conversational data) and write state, chunks, tasks, and topic files to disk; no external network endpoints are contacted by the scripts themselves.
Persistence & Privilege
Skill does not request always:true and does not modify other skills. It writes its own state and outputs to the workspace and memory directories (normal for this functionality). Cron instructions are suggested but are user-controlled (manual crontab edits).
如何使用
- 确保已安装 OpenClaw（本地或 Docker 部署）
- 在对话框中输入安装命令：
/install subagent-distiller - 安装完成后，直接呼叫该 Skill 的名称或使用
/subagent-distiller触发 - 根据 Skill 的参数说明提供必要输入，即可获得结构化输出
版本历史
v3.0.1
- Update version to 3.0.1
- Documentation cleanup in SKILL.md; no user-facing feature changes
- No change in core functions or configuration
v3.0.0
Subagent Distiller v3.0.0 — Major Upgrade
- 全面重构为生产级记忆蒸馏系统，自动提取增量结构化知识
- 新增Cursor增量扫描，显著节省资源，仅处理新增对话内容
- 实时结构化整理、动态聚类，自动发现和维护知识域
- 引入智能过滤机制，自动丢弃短期和无价值内容，专注长期沉淀
- 支持知识卡片状态管理（RESOLVED/PENDING/ABANDONED）与自动提醒
- 新增一键批量清理、域专书聚合等高效管理工具
元数据
常见问题
Subagent Distiller 是什么？
自动增量提取对话中的结构化知识，智能过滤无用信息，动态聚类主题，支持状态追踪和长期价值沉淀。 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件，目前累计下载 366 次。
如何安装 Subagent Distiller？
在 OpenClaw 或 Claude Code 对话框中运行命令「/install subagent-distiller」即可一键安装，无需额外配置。
Subagent Distiller 是免费的吗？
是的，Subagent Distiller 完全免费（开源免费），可自由下载、安装和使用。
Subagent Distiller 支持哪些平台？
Subagent Distiller 跨平台运行，可在任意部署了 OpenClaw / Claude Code 的环境中使用（cross-platform）。
谁开发了 Subagent Distiller？
由 Peng Shu（@mashirops）开发并维护，当前版本 v3.0.1。
推荐 Skills