← 返回 Skills 市场
Extracts durable workflow preferences and project conventions from a session, then proposes the smallest valid update to global or project CLAUDE.md files. Trigger when the user asks to remember rules, when repeated preference patterns appear, or at session end for non-trivial sessions.
作者
TaiChangXieBuWan
· GitHub ↗
· v1.0.3
· MIT-0
136
总下载
0
收藏
0
当前安装
3
版本数
在 OpenClaw 中安装
/install self-improving-session
功能描述
Extracts durable workflow preferences and project conventions from a session, then proposes the smallest valid update to global or project CLAUDE.md files. T...
安全使用建议
This skill appears coherent: it analyzes your current conversation to propose small, high-confidence rules and suggests edits to ~/.claude/CLAUDE.md or a project CLAUDE.md. Before installing or enabling it, consider: (1) the agent will read full sessions to find patterns — avoid putting secrets or sensitive data in the conversation you want analyzed; (2) confirm that file writes require your approval (the SKILL.md says they will), and decline automatic write/automation if you want to review proposed changes; (3) if you prefer, restrict the skill's autonomous invocation or require an explicit user prompt before it runs at session end. If the package later included an automated writer that applied changes without confirmation or added network endpoints, that would change this assessment to suspicious.
功能分析
Type: OpenClaw Skill
Name: self-improving-session
Version: 1.0.3
The 'self-improving-session' skill is a documentation-centric tool designed to help an AI agent extract and persist workflow preferences and project conventions into CLAUDE.md files. It contains no executable code or scripts, relying entirely on instructions (SKILL.md) and reference guidelines (references/learning-rules.md, references/rule-quality.md). The skill includes explicit security safeguards, such as instructions to never store secrets or sensitive data, a requirement for user confirmation before writing to files, and logic to prevent 'over-learning' or bloat. No indicators of data exfiltration, malicious execution, or unauthorized persistence were found.
能力评估
Purpose & Capability
Name/description (extract durable workflow preferences and propose minimal CLAUDE.md updates) matches the SKILL.md behavior: it inspects the session, filters signals, forms concise rule candidates, and targets global or project CLAUDE.md files. There are no unrelated env vars, binaries, or installs requested.
Instruction Scope
Instructions require scanning the current session and optionally events from a paired prompt module; that is appropriate for the stated goal. The skill explicitly forbids storing full prompt text, code facts, secrets, or temporary state and states that writes to CLAUDE.md occur only after user confirmation. This is coherent, but users should note the skill reads conversational content (which may include sensitive data) to extract patterns — the SKILL.md constrains what is recorded but the agent still needs full session access to analyze signals.
Install Mechanism
Instruction-only skill with no install spec, no bundled code, and no external downloads. Lowest-risk install posture; nothing is written to disk by default and no arbitrary code is fetched.
Credentials
No environment variables, credentials, or config paths are required. Requested targets (user global CLAUDE.md and project CLAUDE.md) match the declared purpose.
Persistence & Privilege
always is false and the skill is user-invocable (normal). The skill may propose modifications to CLAUDE.md but states writes require user confirmation; it does not demand permanent privileged presence or automatic file writes in the SKILL.md. Autonomous invocation by the agent is permitted by platform default, which is expected for a skill that monitors sessions.
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 self-improving-session - 安装完成后，直接呼叫该 Skill 的名称或使用
/self-improving-session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
v1.0.3
self-improving-session 1.0.3: Major ruleset and scope tightening
- Greatly clarifies what counts as a durable, reusable session rule versus noise.
- Splits rules by confidence level with clear thresholds for `[tentative]`, `[stable]`, and correction markers.
- Adds explicit rule budget guidance to reduce `CLAUDE.md` bloat (max ~20 global, ~15 project-specific rules).
- Now prefers merging, replacing, or compressing over appending near-duplicates; often proposes zero net change.
- New reference file `references/rule-quality.md` explains what should be learned or skipped.
- Removes embedded session-end automation instructions from core docs (now in README only).
- Stronger, briefer, and safer rule extraction and merge policies.
v1.0.2
Version 1.0.2
- Removed README.md and scripts/self-improving-session-hook.sh, simplifying the skill by eliminating setup and script files.
- Added a **Setup** section in SKILL.md with inline instructions for auto-triggering at session end, replacing information previously found in the removed files.
- All usage and integration guidance is now consolidated in SKILL.md for easier reference.
- No changes to core logic or user-facing functionality.
v1.0.1
- Improves extraction and storage of durable user preferences and rules from each session, aligning future sessions automatically.
- Adds structured process to scan for corrections, repeated workflow patterns, and project conventions, merging them into `CLAUDE.md` with confidence markers (`[tentative]`, `[stable]`).
- Prevents overfitting by filtering out one-off behaviors, prompt text, code facts, and sensitive data.
- Provides clear confirmation and output formatting guidelines for updating rules and handling user corrections.
- Maintains tight integration with `self-improving-prompt` to continuously learn user workflow choices across sessions.
元数据
常见问题
Extracts durable workflow preferences and project conventions from a session, then proposes the smallest valid update to global or project CLAUDE.md files. Trigger when the user asks to remember rules, when repeated preference patterns appear, or at session end for non-trivial sessions. 是什么？
Extracts durable workflow preferences and project conventions from a session, then proposes the smallest valid update to global or project CLAUDE.md files. T... 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件，目前累计下载 136 次。
如何安装 Extracts durable workflow preferences and project conventions from a session, then proposes the smallest valid update to global or project CLAUDE.md files. Trigger when the user asks to remember rules, when repeated preference patterns appear, or at session end for non-trivial sessions.？
在 OpenClaw 或 Claude Code 对话框中运行命令「/install self-improving-session」即可一键安装，无需额外配置。
Extracts durable workflow preferences and project conventions from a session, then proposes the smallest valid update to global or project CLAUDE.md files. Trigger when the user asks to remember rules, when repeated preference patterns appear, or at session end for non-trivial sessions. 是免费的吗？
是的，Extracts durable workflow preferences and project conventions from a session, then proposes the smallest valid update to global or project CLAUDE.md files. Trigger when the user asks to remember rules, when repeated preference patterns appear, or at session end for non-trivial sessions. 完全免费，采用 MIT-0 许可证，可自由下载、安装和使用。
Extracts durable workflow preferences and project conventions from a session, then proposes the smallest valid update to global or project CLAUDE.md files. Trigger when the user asks to remember rules, when repeated preference patterns appear, or at session end for non-trivial sessions. 支持哪些平台？
Extracts durable workflow preferences and project conventions from a session, then proposes the smallest valid update to global or project CLAUDE.md files. Trigger when the user asks to remember rules, when repeated preference patterns appear, or at session end for non-trivial sessions. 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 Extracts durable workflow preferences and project conventions from a session, then proposes the smallest valid update to global or project CLAUDE.md files. Trigger when the user asks to remember rules, when repeated preference patterns appear, or at session end for non-trivial sessions.？
由 TaiChangXieBuWan（@lq434239）开发并维护，当前版本 v1.0.3。
推荐 Skills