📦 Context Compression — 技能工具
v3.13.3Prevent context overflow with automatic session truncation and memory preservation.
3· 1.6k·23 当前·23 累计
安全扫描
OpenClaw
安全
high confidenceThe skill's code and runtime instructions are consistent with its stated purpose (truncating session files and preserving extracted facts) and request only local OpenClaw workspace access; no unrelated credentials or external downloads are required.
评估建议
This package appears to do what it says: periodically trim session JSONL files and preserve important lines into MEMORY.md/daily notes. Before installing: 1) Backup ~/.openclaw/workspace/MEMORY.md and your sessions directory. 2) Review/adjust the cron path printed by configure.sh — configure.sh expects a 'scripts' subdirectory but the packaged scripts live in the skill root; ensure your cron points to the actual truncate-sessions-safe.sh location. 3) If you enable AI-assisted identification, be ...详细分析 ▾
✓ 用途与能力
Name/description match the actual behavior: scripts read OpenClaw session JSONL files, score/prioritize lines, truncate sessions, and append identified facts to MEMORY.md and daily notes. The declared binaries and file paths align with this function. The one minor mismatch: some installation/docs reference scripts placed directly in ~/.openclaw/workspace/skills/context-compression/, while configure.sh writes/prints paths under a 'scripts' subdirectory ($WORKSPACE/skills/context-compression/scripts) — this is likely a packaging/installation path bug, not a capability mismatch.
ℹ 指令范围
Runtime instructions and scripts operate on local OpenClaw files (sessions, MEMORY.md, memory/). The optional AI-assisted fact extractor invokes the local 'openclaw agent' CLI and passes conversation content to it — that CLI may, depending on the user's OpenClaw configuration, forward data to remote LLMs. The SKILL.md notes this, but users should be aware that enabling AI-assisted identification will cause conversation content to leave the local truncation scripts via the configured OpenClaw agent pathway. Otherwise, the scripts do not perform external HTTP calls themselves.
✓ 安装机制
No install spec / no external downloads are required (instruction-only skill with bundled scripts). Files are intended to be placed in the standard OpenClaw skill path; nothing in the package auto-downloads code from untrusted URLs or writes to system-wide nonstandard locations.
✓ 凭证需求
The skill requires only standard shell utilities and reads/writes files under the user's OpenClaw workspace (~/.openclaw). It does not request environment variables or external service credentials. The only indirect network risk is via the optional call to the local 'openclaw' CLI, which depends on the user's OpenClaw/network configuration — this is proportional but should be noted by privacy-conscious users.
✓ 持久化与权限
The skill does not request always:true and is designed to be invoked by cron/hooks or manually. configure.sh prompts the user to add cron entries rather than modifying crontab silently. Scripts write only to their own workspace files (config, MEMORY.md, memory/), and do not modify other skills' configurations.
安全有层次,运行前请审查代码。
运行时依赖
无特殊依赖
版本
latestv3.13.32026/3/8
Simplify post-install callout to plain text
● 无害
安装命令
点击复制官方npx clawhub@latest install context-compression
镜像加速npx clawhub@latest install context-compression --registry https://cn.longxiaskill.com
技能文档
Keep conversations within limits. Never lose important context.
⚡ After installing, run the interactive setup wizard to generate your config file, then add the suggested cron entry with crontab -e. See Quick Start below for commands.Quick Start
File location: ClawHub installs this skill to ~/.openclaw/workspace/skills/context-compression/. All scripts are placed here directly. This is the standard OpenClaw skill install path — no manual file placement needed.# 1. Install and configure (interactive) bash ~/.openclaw/workspace/skills/context-compression/configure.sh# 2. Verify config exists cat ~/.openclaw/workspace/.context-compression-config.json
# 3. Set up crontab (example: every 10 minutes) /10 ~/.openclaw/workspace/skills/context-compression/truncate-sessions-safe.sh
How It Works
Session Truncation (truncate-sessions-safe.sh)
- Scheduling: System crontab (e.g.,
/10) - Action: Reads
.jsonlsession files under~/.openclaw/agents/*/sessions/, trims each file to the configured size - Safety: Skips files with a matching
.lockfile (active session) - Integrity: Keeps JSONL line boundaries intact — never splits a line
- Strategy:
priority-firstscans for important keywords before trimming and preserves matching lines
Fact Identification
- Keyword-based:
identify-facts.sh— scans truncated content for keywords (重要, 决定, TODO, 偏好, deadline, must remember, etc.) and appends findings tomemory/YYYY-MM-DD.md - AI-assisted:
identify-facts-enhanced.sh— callsopenclaw agent --agent main --messagewith the trimmed content to semantically identify important facts. Only used whenopenclawCLI is available on PATH. - Triggered by:
truncate-sessions-safe.shcalls one of these before each truncation cycle
Preference Lifecycle (check-preferences-expiry.sh)
- Scheduling: Once daily via crontab
- Mechanism: Reads MEMORY.md preference entries tagged with
@YYYY-MM-DD, removes expired ones - Tiers: Short-term (1-7 days), Mid-term (1-4 weeks), Long-term (permanent)
Scripts
| Script | Purpose | Scheduling |
|---|---|---|
truncate-sessions-safe.sh | Trim session JSONL files | crontab, every 10 min |
identify-facts.sh | Keyword-based fact detection | Called by truncate script |
identify-facts-enhanced.sh | AI-assisted fact detection | Called by truncate script |
check-preferences-expiry.sh | Remove expired preferences | crontab, once daily |
configure.sh | Interactive setup wizard | Manual, one-time |
session-start-hook.sh | Load context at session start | Called by AGENTS.md |
session-end-hook.sh | Save context at session end | Called by AGENTS.md |
check-context-health.sh | Report current context status | Manual / on-demand |
Configuration
File: ~/.openclaw/workspace/.context-compression-config.json
{
"version": "2.3",
"maxChars": 40000,
"frequencyMinutes": 10,
"skipActive": true,
"strategy": "priority-first",
"useAiIdentification": false,
"priorityKeywords": [
"重要", "决定", "记住", "TODO", "偏好",
"important", "remember", "must", "deadline"
]
}
| Field | Type | Default | Description |
|---|---|---|---|
| maxChars | number | 40000 | Max chars to keep per session file |
| frequencyMinutes | number | 10 | How often crontab runs truncate |
| skipActive | boolean | true | Skip sessions with .lock files |
| strategy | string | priority-first | Truncation strategy |
| useAiIdentification | boolean | false | Set true to use AI-assisted fact identification (may send content to remote LLMs) |
| priorityKeywords | string[] | (see above) | Keywords to preserve during truncation |
Troubleshooting
| Problem | Solution |
|---|---|
| Context still exceeded | Reduce maxChars in config |
| Memory not persisting | Check that AGENTS.md includes session-start-hook |
| Crontab not running | Verify PATH in crontab includes node/openclaw binary location |
Safety
Data Protection
- No deletion: Truncation writes the trimmed portion back to the same file. It does not delete files.
- Backup before trim:
truncate-sessions-safe.shcreates a.pre-trimbackup of each file before modification. Backups are cleaned up after a successful write. - Line integrity: Truncation only cuts at JSONL line boundaries. Partial lines are never written.
- Active sessions protected: Files with a matching
.lock(currently in use) are always skipped, even if oversized.
Safe Defaults
All configuration values have conservative defaults:skipActive: true— never touches a running sessionmaxChars: 40000— keeps substantial history per sessionstrategy: priority-first— preserves lines matching priority keywords before trimming anything- No direct network access from scripts. The optional AI fact identification uses your local
openclawCLI — network activity depends on your OpenClaw configuration.
User Control
- Crontab: The user creates and manages all scheduled tasks. No script auto-installs crontab entries.
- Configuration: All settings live in a single JSON file. The
configure.shwizard runs interactively and requires user input. - Opt-out: Remove the crontab entry to stop all automated truncation. The skill has no background daemon of its own.
- Scope: Only reads/writes files under
~/.openclaw/agents/main/and~/.openclaw/workspace/memory/. Never touches system files, other agents' data, or other users' data.
Privacy Notice
- AI-assisted fact identification (
identify-facts-enhanced.sh) is disabled by default. It invokes the localopenclaw agentCLI, which may send session content to remote LLM services depending on your OpenClaw configuration. Only enable it if you understand and accept this data flow. To enable, set"useAiIdentification": truein the config file. - Keyword-based identification (
identify-facts.sh) is the default and runs entirely locally with no external data transmission. - Unattended cron execution: If you enable cron jobs, the scripts run without interactive consent. Review the scripts and test manually before enabling scheduled runs.