OpenClaw: compact improver

优化 OpenClaw 4.2 的 /compact 压缩命令提示词，生成结构化摘要。当用户提到压缩效果差、摘要不完整、上下文丢失、/compact 格式不好，或要求改进压缩提示词时触发。

This skill intends to improve OpenClaw's /compact prompt and is instruction-only, but review these before installing or using it: - Source/trust: The skill has no homepage and an unknown source — verify the author before allowing it to modify your agent config. - Config injection: SKILL.md tells the agent to inject a new systemPrompt into ~/.openclaw/openclaw.json. The registry metadata did not declare any config-path access; do not let it modify your config without an explicit, reviewed change and a backup. - File exposure: The prompts request preserving exact file paths and full runnable code snippets and instruct the agent to grep and verify files. That can expose source code and secrets. Only enable this behavior when you explicitly want code included and after confirming which directories are in scope. - Contradictory guidance: The files say both 'do not use tools' and to run grep — clarify with the author which actions the skill will actually perform. Recommendations: - Ask the author to update the manifest to declare the config-path and justify file access, or provide a non-persistent workflow (manual prompt you paste into the UI). - If you test it, do so on a sandboxed project or with a backup of ~/.openclaw/openclaw.json. - Limit the agent's filesystem access or restrict which directories it may read to avoid accidental exfiltration of sensitive code/config. Given these mismatches (undeclared config modification and broad file-reading instructions), treat the skill as suspicious until the author clarifies and documents the exact behaviors and required permissions.

Type: OpenClaw Skill Name: openclaw-compact-improver Version: 1.0.0 The skill bundle is a prompt-engineering utility designed to improve the quality of OpenClaw's '/compact' command by implementing a structured 9-part summary format. It contains no executable code, exfiltration logic, or malicious instructions; instead, it provides templates and 'trust rules' (e.g., verifying file paths via grep) to ensure the agent maintains technical accuracy and context during conversation compression (SKILL.md, references/compact-prompt.md).