← 返回 Skills 市场
Self-Improving Conversation
作者
José I. O.
· GitHub ↗
· v1.1.0
· MIT-0
99
总下载
0
收藏
0
当前安装
2
版本数
在 OpenClaw 中安装
/install self-improving-conversation
功能描述
Captures dialogue learnings, tone mismatches, escalation failures, and conversation quality issues for continuous improvement. Use when: (1) A user expresses...
安全使用建议
This skill appears coherent for its stated purpose. Before installing: (1) review the shell scripts (activator, error-detector, extract-skill) to confirm you accept local file writes (they create .learnings/ and may scaffold skills under ./skills or ~/.openclaw); (2) only enable the optional OpenClaw hooks if you trust the source—hooks inject a reminder on every agent bootstrap and run with the agent's permissions; (3) note the error-detector reads CLAUDE_TOOL_OUTPUT (platform-provided) which can contain sensitive output—ensure your environment won't leak secrets into that variable; (4) verify the GitHub repo and author if you plan to clone automatically. If you want lower risk, use the scripts manually or keep the hook disabled.
功能分析
Type: OpenClaw Skill
Name: self-improving-conversation
Version: 1.1.0
The skill bundle is a framework for AI agents to track and improve conversational quality by logging issues like tone mismatches and hallucinations to local markdown files. It utilizes standard shell scripts (e.g., `activator.sh`, `extract-skill.sh`) and OpenClaw hooks to manage these logs and promote recurring patterns into project memory. The bundle contains no evidence of data exfiltration, malicious execution, or harmful prompt injection, and it explicitly instructs the agent to avoid logging personally identifiable information or authentication tokens.
能力评估
Purpose & Capability
The name/description (capture dialogue learnings and produce skill scaffolds) match the provided files: markdown templates, scripts to create .learnings entries and to scaffold skills, and OpenClaw hook handlers that inject a reminder. No unrelated credentials, binaries, or network-only installers are requested.
Instruction Scope
Runtime instructions and scripts create and update .learnings/* files in the workspace (or ~/.openclaw workspace when following OpenClaw guidance). The error detector reads the CLAUDE_TOOL_OUTPUT environment variable (a platform-provided output variable) to detect patterns; SKILL.md explicitly warns about not logging PII. The scripts will write files to disk when run (extract-skill.sh, activator, etc.) and the hook injects a virtual reminder file at bootstrap—these behaviours are within the stated scope but will modify local workspace state if enabled.
Install Mechanism
There is no opaque network install: recommended installs are via clawdhub or a git clone from GitHub. No downloads from untrusted shorteners or personal IP addresses are used. There is no install spec that silently fetches/extracts arbitrary archives.
Credentials
The skill declares no required env vars or credentials. Scripts reference CLAUDE_TOOL_OUTPUT (used by the PostToolUse hook) and event objects in hook handlers—these are platform integration values rather than secrets. No sensitive credentials (API keys, AWS, tokens) are requested. Be aware that CLAUDE_TOOL_OUTPUT may contain sensitive tool output depending on your configuration; the docs even warn not to forward it verbatim.
Persistence & Privilege
always is false and the skill is opt-in. Hook behaviour is optional: enabling hooks copies files into ~/.openclaw/hooks or similar and the handler injects a virtual reminder into bootstrapFiles. The skill does not attempt to change other skills' configs or escalate privileges.
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-conversation - 安装完成后，直接呼叫该 Skill 的名称或使用
/self-improving-conversation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
v1.1.0
**Version 1.1.0**
- Added stackability contract for multi-skill installations.
- Added namespaced logging guidance (`.learnings/conversation/`) for coexistence with other skills.
- Added required `Skill: conversation` metadata field and cross-skill precedence/ownership rules.
- Clarified hook arbitration model (single dispatcher, dedupe, rate limiting).
v1.0.0
- Initial release of the self-improving-conversation skill.
- Captures dialogue learnings, tone mismatches, escalation failures, and conversation quality issues to markdown logs for continuous improvement.
- Introduces structured categories like tone_mismatch, misunderstanding, escalation_failure, context_loss, sentiment_drift, and hallucination.
- Provides setup instructions and logging templates for both OpenClaw and generic agents.
- Includes workflow guidance for reviewing issues and promoting key patterns to project memory files.
- Emphasizes privacy by avoiding logging of sensitive user data.
元数据
常见问题
Self-Improving Conversation 是什么？
Captures dialogue learnings, tone mismatches, escalation failures, and conversation quality issues for continuous improvement. Use when: (1) A user expresses... 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件，目前累计下载 99 次。
如何安装 Self-Improving Conversation？
在 OpenClaw 或 Claude Code 对话框中运行命令「/install self-improving-conversation」即可一键安装，无需额外配置。
Self-Improving Conversation 是免费的吗？
是的，Self-Improving Conversation 完全免费，采用 MIT-0 许可证，可自由下载、安装和使用。
Self-Improving Conversation 支持哪些平台？
Self-Improving Conversation 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 Self-Improving Conversation？
由 José I. O.（@jose-compu）开发并维护，当前版本 v1.1.0。
推荐 Skills