Compact Test A

Smart context compaction for OpenClaw agents. 4-phase progressive strategy: Scan, Extract, Check, Compact. Before running /compact, this skill scans tool out...

Security Scan

The skill's stated purpose (pre-compact scanning and saving important outputs to memory files) matches its instructions, but the runtime instructions ask the agent to scan and persist tool outputs (which can contain secrets) without a technically enforceable redaction mechanism — this creates a risk of sensitive data being written to persistent memory.

Purpose & Capability

Name/description (Smart Compact) align with the instructions: it scans tool outputs, extracts important facts, writes them into memory/YYYY-MM-DD.md, and generates a checklist before running /compact. No unrelated binaries, env vars, or install steps are requested.

Instruction Scope

SKILL.md explicitly instructs the agent to review all tool outputs (exec, read, web_fetch, web_search) and extract items such as IPs, endpoints, file paths, and even 'authentication information' (albeit claiming it will be redacted). This gives the agent broad discretion to persist potentially sensitive data from tool outputs to disk. There is no concrete, auditable redaction algorithm or enforcement mechanism in the instructions — it's a policy statement the agent is asked to follow, not a code-level guarantee.

Install Mechanism

This is an instruction-only skill in the registry (no install spec). README offers cloning or curl-from-GitHub as optional installation methods; those are common but carry the usual risk of pulling remote content. Nothing in the registry forces an arbitrary binary download or execution.

Credentials

The skill declares no required env vars or credentials, which is consistent. However, its operation depends on reading tool outputs and writing memory files; if those outputs include credentials or tokens, they may be persisted. The skill does not request unrelated external credentials, but it effectively needs access to agent tool outputs and file write permission (memory directory) — reasonable for the stated purpose but potentially sensitive.

Persistence & Privilege

The skill does not request always:true and does not change other skills. It explicitly writes persistent memory files (memory/YYYY-MM-DD.md) which is expected for its purpose. Persisting arbitrary extracted content (including secrets if they appear in outputs) is the main persistence-related risk and relies on policy-level redaction rather than enforced safeguards.

What to consider before installing

What to consider before installing: - Understand what 'memory/YYYY-MM-DD.md' path refers to on your system and who can read it; inspect and set file permissions after installation. - The skill will scan tool outputs (exec/read/web_fetch/web_search). Those outputs can contain secrets (API keys, tokens, private file paths). Ask: do you want agent-written files to potentially include such data? - The SKILL.md promises redaction of sensitive info but provides no technical guarantee. If you rely on this, test with non-sensitive examples and review the produced memory files to verify redaction behavior. - If you install from GitHub/raw URLs, review the repository contents (README and SKILL.md are visible here) and prefer a vetted source; consider cloning from a trusted repo or verifying commit hashes. - If you want to reduce risk: restrict the agent's tools so it cannot read sensitive files or environment variables, require explicit user confirmation for each memory write, or run Smart Compact in a sandbox/with logs reviewed manually. - Periodically audit memory files and consider an automatic cleanup or encryption policy for stored memories that may contain sensitive items.

Smart Compact — 智能压缩增强

四阶段渐进式压缩策略，在 /compact 前先把重要信息救出来。

什么时候用

• 用户说"智能压缩"、"smart-compact"、"压缩检查"
• 在手动执行 /compact 之前先跑一遍
• 对话上下文快满时，主动触发
• Heartbeat 检测到 context 接近 80% 时自动建议

核心理念

传统的上下文压缩是一刀切——整个对话被浓缩成一段摘要，大量细节在过程中丢失。

Smart Compact 采用四阶段渐进式策略，在 /compact 之前插入一个"预处理"阶段：

1. 扫描：识别对话中的大块工具输出和关键信息
2. 提取：把值得保留的信息写入记忆文件
3. 检查：生成压缩前检查清单，标记风险项
4. 压缩：用户确认安全后才执行压缩

核心原则：先救再压，宁可多存也不能漏存。

执行流程

Phase 1 — 扫描工具输出

1. 回顾当前对话中所有的工具调用结果
2. 识别大块输出（超过 50 行或 2000 字符的工具结果）
3. 对每个大块输出评估：
   • 是否包含关键信息（决策、配置、错误信息、地址等）
   • 是否已经被后续对话引用或总结过
   • 是否是重复或冗余的（如多次 ls、git status）

Phase 2 — 提取记忆

1. 从工具输出和对话中提取值得持久化的信息：

   • 新发现的事实：地址、配置值、端点、文件路径
   • 决策和原因：为什么选了方案 A 而不是 B
   • 错误和解决方案：踩坑记录
   • 用户偏好：明确表达的喜好或要求
   • 任务进度：哪些做完了，哪些还没做

2. 将提取的信息追加写入 memory/YYYY-MM-DD.md

   • 使用 edit（追加模式），绝不覆盖已有内容
   • 每条记忆附带简短的来源说明

Phase 3 — 生成压缩前检查清单

输出一份结构化的检查清单：

📋 Smart Compact 检查清单
━━━━━━━━━━━━━━━━━━━━━━

📊 扫描统计：
- 工具调用总数：N 次
- 大块输出（>50行）：N 个
- 已引用/总结过的：N 个
- 可安全压缩的：N 个

💾 已提取到记忆：
- [+] 新事实：简要描述...
- [+] 决策记录：简要描述...
- [+] 错误解决：简要描述...
（共 N 条写入 memory/YYYY-MM-DD.md）

⚠️ 需要注意：
- [!] 某某工具输出包含重要数据但尚未被引用
- [!] 某某配置值只出现在工具输出中

✅ 建议：可以安全执行 /compact

Phase 4 — 执行压缩（可选）

• 如果检查清单显示"✅ 可以安全压缩"，提示用户确认
• 用户确认后，执行 /compact
• 如果有 ⚠️ 警告项，先处理完再压缩

规则

必须遵守

• 绝不丢弃未被记录的关键信息：宁可多存也不能漏存
• 追加写入：只用 edit 追加到 memory 文件，绝不覆盖
• 不自动压缩：除非用户明确确认，否则只生成检查清单
• 透明：每一步操作都告知用户

信息分类标准

• 必须保存：重要配置、地址端点、文件路径、错误解决方案
• 建议保存：决策原因、用户偏好、任务进度
• 可以丢弃：重复的 ls 输出、已被总结的搜索结果、中间调试过程

与 Dream Skill 的配合

Smart Compact 和 Dream 是互补的：

• Smart Compact：实时的，在压缩前抢救信息 → 写入日记
• Dream：定期的，把日记整合到长期记忆 → 更新 MEMORY.md

推荐工作流：

1. 对话中随时触发 Smart Compact 保护信息
2. 每天凌晨 Dream 整合日记到长期记忆
3. 形成完整的记忆保护链条