context-saver

分类: 工具与效率 | 上传者: tlancas25 | 下载: 0 | 版本: v1.0（最新）

OpenClaw v2.0 的令牌保存执行层。在沙盒子进程中运行技能命令，其中只有紧凑的摘要进入上下文窗口。通过 SQLite 事件跟踪提供会话连续性，可在对话压缩中幸存下来。支持意图驱动的过滤、批量多技能执行和渐进式内存加载。包括自动安装程序，可通过单个命令将上下文保护程序连接到 AGENTS.md、TOOLS.md 和 cron 作业中。使用此技能包装任何数据量大的操作，以减少 70-98% 的令牌消耗。

更新日志: Source: GitHub https://github.com/tlancas25/openclaw-context-saver

目录结构

当前层级: tree/main/

📁 docs/
- 📄 ARCHITECTURE.md 8.6 KB
- 📄 BENCHMARKS.md 6.4 KB
- 📄 INTEGRATION.md 7.8 KB
- 📄 savings-breakdown.png 1.4 MB
- 📄 savings-chart.png 226.8 KB
📁 examples/
- 📄 daily-status-pipeline.json 1.3 KB
- 📄 eod-report-pipeline.json 1.7 KB
📁 scripts/
- 📄 ctx_batch.py 5.6 KB
- 📄 ctx_run.py 15.9 KB
- 📄 ctx_search.py 4.0 KB
- 📄 ctx_session.py 9.5 KB
- 📄 ctx_stats.py 4.4 KB
- 📄 deliver.py 8.3 KB
- 📄 morning_brief_pipeline.py 9.5 KB
📁 src/
- 📁 lib/
  - 📄 chunker.ts 3.1 KB
  - 📄 db.ts 6.4 KB
  - 📄 env.ts 1.8 KB
  - 📄 filter.ts 5.4 KB
  - 📄 redact.ts 1.1 KB
  - 📄 sandbox.ts 6.0 KB
- 📁 tools/
  - 📄 batch.ts 4.0 KB
  - 📄 deliver.ts 5.5 KB
  - 📄 doctor.ts 5.6 KB
  - 📄 execute-file.ts 3.7 KB
  - 📄 execute.ts 5.3 KB
  - 📄 fetch-index.ts 3.5 KB
  - 📄 index.ts 2.1 KB
  - 📄 search.ts 1.4 KB
  - 📄 session.ts 6.8 KB
  - 📄 stats.ts 3.6 KB
- 📄 server.ts 8.9 KB
- 📄 types.ts 2.3 KB
📄 .gitignore 77 B
📄 install.py 34.8 KB
📄 LICENSE 1.1 KB
📄 package-lock.json 57.2 KB
📄 package.json 1.0 KB
📄 README.md 15.6 KB
📄 skill.json 1.0 KB
📄 SKILL.md 3.5 KB
📄 tsconfig.json 452 B

SKILL.md

---
name: context-saver
description: "Token-saving execution layer for OpenClaw v2.0. Runs skill commands in sandboxed subprocesses where only compact summaries enter the context window. Provides session continuity via SQLite event tracking that survives conversation compaction. Supports intent-driven filtering, batched multi-skill execution, and progressive memory loading. Includes automated installer that wires context-saver into AGENTS.md, TOOLS.md, and cron jobs with a single command. Use this skill to wrap any data-heavy operation to reduce token consumption by 70-98%."
metadata: {"openclaw":{"emoji":"🪶","requires":{"bins":["python3"],"env":[]}}}
---

# Context Saver — Token Optimization for OpenClaw

Reduces token consumption by 70-98% through sandboxed execution, intent filtering, and session continuity.

## Core Concept

Instead of dumping raw API responses (3-50 KB) into context, Context Saver:
1. Runs skill commands in subprocesses
2. Captures full output
3. Extracts only relevant fields based on intent
4. Returns compact summaries (100-500 bytes)
5. Indexes full output in FTS5 for on-demand retrieval

## Commands

### Sandboxed Skill Execution
```bash
# Run any skill command with automatic summarization
python3 scripts/ctx_run.py --skill my-api --cmd "dashboard" [--intent "check error rate"]
python3 scripts/ctx_run.py --skill my-api --cmd "list-items" [--intent "find failures"]
python3 scripts/ctx_run.py --skill my-api --cmd "search users" --intent "inactive accounts over 90 days"

# Run with explicit summary fields
python3 scripts/ctx_run.py --skill my-api --cmd "dashboard" --fields "active_users,error_rate,uptime"
```

### Batch Execution (Multiple Skills in One Call)
```bash
python3 scripts/ctx_batch.py --commands '[
  {"skill": "my-api", "cmd": "dashboard", "fields": ["active_users","error_rate"]},
  {"skill": "analytics-engine", "cmd": "metrics", "intent": "summary"},
  {"skill": "health-monitor", "cmd": "check", "intent": "failures only"}
]'
```

### Session Event Tracking
```bash
# Log a session event
python3 scripts/ctx_session.py log --type "action" --priority critical --data '{"operation":"deploy","service":"api-v2"}'

# Build compaction snapshot (called before conversation compacts)
python3 scripts/ctx_session.py snapshot

# Restore from snapshot (called on session resume)
python3 scripts/ctx_session.py restore

# View session stats
python3 scripts/ctx_session.py stats
```

### Context Search (Query Indexed Data)
```bash
python3 scripts/ctx_search.py "error rate spike" [--source my-api]
python3 scripts/ctx_search.py "failed deployments" --source last-run
```

### Context Stats
```bash
python3 scripts/ctx_stats.py
# Shows: total bytes saved, calls made, avg compression ratio, session events
```

## How It Saves Tokens

| Operation | Without Context Saver | With Context Saver | Savings |
|-----------|----------------------|-------------------|---------|
| Dashboard summary | 3 KB JSON | 120 B summary | 96% |
| List items (50 records) | 5 KB JSON | 300 B summary | 94% |
| Search results (200 hits) | 20-50 KB | 500 B filtered | 97% |
| Multi-skill pipeline | 23 KB (4 calls) | 2 KB (1 batch) | 91% |
| Session cold start | 20 KB workspace files | 2 KB snapshot | 90% |

## When to Use
- Any data-heavy skill command (APIs, analytics, search)
- Multi-step pipelines (daily reports, status checks)
- Session resumption after compaction
- When context window is running low
- Don't use for small operations (<500 bytes output)
- Don't use when you need full raw data for editing/modification

登录后下载/点赞/收藏 ❤ 10 | ★ 0

context-saver

目录结构

SKILL.md

举报内容

提示