← Back to Skills Marketplace
持久化状态格式（
配置文件（
环境变量（与
236
Downloads
0
Stars
0
Active Installs
2
Versions
Install in OpenClaw
/install context-switch-token-optimizer
Description
智能对话上下文切换与 Token 优化。分析对话主题连续性以管理上下文、在话题间切换并节省 Token。 适用于多任务对话、项目工作流、学习研究等需管理上下文与 Token 使用的场景。
README (SKILL.md)
Context Switch Token Optimizer
技能描述
通过关键词相似度、弱关联子串与时间衰减判断「连续 / 渐变漂移压缩 / 硬切换」，并可监控 Token、触发压缩与清理。实现见 context_manager.py、token_optimizer.py；详细算法见 docs/TOPIC_AND_CONTINUITY.md。
上下文动作（与代码一致）
| 动作 | 条件 | 行为 |
|---|---|---|
load_memory |
无当前主题且无历史 | 建主题、可选加载 memory/topic 下记忆 |
continuous |
与上一轮相似度 ≥ similarity_threshold |
追加历史，不压缩 |
drift_compress |
相似度 ∈ [continuity_threshold, similarity_threshold) |
对话仍连续；压缩与当前弱相关的历史轮（is_compressed，保留短摘要） |
switch_context |
相似度 \x3C continuity_threshold |
硬切换，换记忆、增 switch_count |
相似度计算：calculate_topic_similarity()（关键词 Jaccard 式重叠 × 时间衰减 + 2 字子串/首字弱关联 + continuity_bonus_for_partial，弱关联封顶 drift_similarity_cap）。
持久化状态格式（memory/context_switch_state.json）
与 ContextState / TopicSummary 一致，主题历史项包含：
topic,keywords,timestamp,tokens_used,content_snippet,is_compressed- 顶层另有：
current_topic,memory_context,last_switch_time,switch_count,total_tokens
配置文件（config.json）
所有键名与 load_skill_config() 默认值一致；用户 JSON 按顶层块与默认合并（同一块内键覆盖，未写的子键保留默认）。
context_switch
| 键名 | 类型 | 默认 | 说明 |
|---|---|---|---|
similarity_threshold |
number | 0.7 | ≥ 此值判为 continuous |
continuity_threshold |
number | 0.35 | \x3C 此值判为硬切换 switch_context |
compress_relevance_threshold |
number | 0.3 | drift_compress 时，历史轮与当前相似度低于此则压缩该轮 |
continuity_bonus_for_partial |
number | 0.3 | 存在弱关联时的相似度加成 |
drift_similarity_cap |
number | 0.65 | 弱关联时相似度上限，便于进入 drift_compress |
time_decay_factor |
number | 0.95 | 按小时衰减的底数 |
max_topic_history |
int | 10 | 主题历史条数上限 |
memory_relevance_threshold |
number | 0.5 | 记忆文件匹配阈值 |
log_level |
string | （不设） | 可由环境变量 CONTEXT_SWITCH_LOG_LEVEL 写入 |
token_optimization
| 键名 | 类型 | 默认 | 说明 |
|---|---|---|---|
enabled |
bool | true | 为 false 时不自动触发优化（对应环境变量 TOKEN_OPTIMIZER_ENABLED） |
token_limit |
int | 80000 | Token 预算 |
compression_threshold |
int | 56000 | 压缩参考阈值 |
context_cleanup_threshold |
number | 0.8 | 上下文清理参考 |
memory_load_limit |
int | 2000 | 记忆加载参考 |
optimization_interval |
int | 300 | 优化间隔（秒，预留） |
max_history_size |
int | 100 | Token 使用记录条数上限 |
memory_search
| 键名 | 类型 | 默认 | 说明 |
|---|---|---|---|
max_search_results |
int | 5 | 最多返回记忆条数 |
keyword_limit |
int | 5 | 每轮提取关键词数 |
search_depth |
int | 2 | 可由环境变量 MEMORY_SEARCH_DEPTH 覆盖（1–3） |
file_types |
string[] | ["*.md"] |
记忆目录 glob |
optimization_settings
| 键名 | 说明 |
|---|---|
auto_optimization |
是否自动优化（策略预留） |
health_score_threshold |
健康分阈值 |
suggestions_check_interval |
建议检查间隔（秒） |
环境变量（与 apply_environment_overrides() 一致）
| 环境变量 | 映射配置键 | 取值与行为 |
|---|---|---|
CONTEXT_HISTORY_SIZE |
context_switch.max_topic_history |
正整数，限制在 1–100 |
MEMORY_SEARCH_DEPTH |
memory_search.search_depth |
正整数，限制在 1–3 |
TOKEN_OPTIMIZER_ENABLED |
token_optimization.enabled |
true/1/yes/on 为启用；false/0/no/off 为禁用（禁用后使用率 >70% 也不触发 trigger_optimization） |
CONTEXT_SWITCH_LOG_LEVEL |
context_switch.log_level |
DEBUG / INFO / WARNING / ERROR / CRITICAL；设置根 logger 级别 |
未设置上述环境变量时，行为完全由 config.json 与代码默认决定。
CLI 与入口
- 主入口：
python3 context-switch-token-optimizer.py [--config path] --process|--status|--reset|--test - 子模块：
context_manager.py、token_optimizer.py（参数见--help）
维护: Hermosa
版本: v1.1
最后更新: 2026-03-17
Usage Guidance
This skill appears internally consistent and implements the described context-switching and token-optimization behavior without requesting secrets or performing network calls. Before installing/running: 1) Inspect or run it in an isolated directory (it will create/modify memory/context_switch_state.json and other files). 2) If you don't want automatic optimization, set TOKEN_OPTIMIZER_ENABLED=false to disable auto-triggering. 3) Enable debugging (CONTEXT_SWITCH_LOG_LEVEL=DEBUG) if you want to observe similarity calculations and compression decisions. 4) Review the memory/ files after test runs and confirm behavior on representative conversations. If you need the skill to never write to disk, do not run it or run with a sandboxed working directory.
Capability Analysis
Type: OpenClaw Skill
Name: context-switch-token-optimizer
Version: 1.0.1
The skill bundle implements a legitimate utility for managing LLM conversation context and optimizing token usage through topic similarity analysis and history compression. The code operates locally on state files within a designated 'memory/' directory, lacks network communication capabilities, and contains no evidence of malicious execution (e.g., eval/exec), data exfiltration, or prompt injection attacks. All logic in 'context_manager.py' and 'token_optimizer.py' is consistent with the stated purpose of maintaining conversation continuity while reducing token overhead.
Capability Assessment
Purpose & Capability
Name/description (context switching and token optimization) match the code and SKILL.md. The skill only requests configuration and logging environment overrides relevant to topic management and token optimization; no unrelated credentials, binaries, or external services are required.
Instruction Scope
SKILL.md and CLI instruct running the included Python scripts which read/write a local state file (memory/context_switch_state.json), load config.json, and may scan local memory files (configured via file_types). The runtime instructions do not ask the agent to read arbitrary OS secrets or network endpoints. Note: the skill will create and modify files in the working directory (memory/), so consider where the agent runs.
Install Mechanism
There is no install spec (instruction-only from registry perspective), which is low-risk. However the package includes multiple Python source files (no external downloads or package installers). Because code is present, running the skill will write state to disk; no remote fetch/install behavior was observed.
Credentials
Environment variables the skill reads (CONTEXT_HISTORY_SIZE, MEMORY_SEARCH_DEPTH, TOKEN_OPTIMIZER_ENABLED, CONTEXT_SWITCH_LOG_LEVEL) are directly tied to its configuration and behavior. No secrets/keys/tokens/passwords are required.
Persistence & Privilege
The skill persists its own state under memory/context_switch_state.json and updates it during operations (save_context_state). always:false and normal autonomous invocation are set. This is expected for a context manager, but be aware it will create/modify files in the agent's working directory.
How to Use
- Make sure OpenClaw is installed (local or Docker)
- Run the install command in chat:
/install context-switch-token-optimizer - After installation, invoke the skill by name or use
/context-switch-token-optimizer - Provide required inputs per the skill's parameter spec and get structured output
Version History
v1.0.1
- Added initial implementation files: context management and token optimization modules, plus test and debug scripts.
- Provided detailed algorithm and configuration documentation.
- Introduced a richer state persistence format including compression status and token usage tracking.
- Supported advanced topic similarity calculation (Jaccard overlap, substring/partial bonus, time decay).
- Environment variable overrides and merged user config supported.
- Initial CLI support for processing, status, reset, and testing operations.
v1.0.0
Initial release of context-switch-token-optimizer:
- Introduces intelligent context switching, summarizing dialog themes and managing conversation continuity.
- Implements a three-level topic similarity mechanism for continuous, drift-compress, or hard context switching.
- Features topic extraction, keyword analysis, and related memory search for efficient context management.
- Includes token optimization strategies with progressive memory loading and context cleanup.
- Provides monitoring and reporting of token usage and topic switches.
- Suitable for workflows requiring efficient context and token management across conversations.
Metadata
Frequently Asked Questions
What is Context Switch Token Optimizer?
智能对话上下文切换与 Token 优化。分析对话主题连续性以管理上下文、在话题间切换并节省 Token。 适用于多任务对话、项目工作流、学习研究等需管理上下文与 Token 使用的场景。 It is an AI Agent Skill for Claude Code / OpenClaw, with 236 downloads so far.
How do I install Context Switch Token Optimizer?
Run "/install context-switch-token-optimizer" in the OpenClaw or Claude Code chat to install it in one step — no extra setup required.
Is Context Switch Token Optimizer free?
Yes, Context Switch Token Optimizer is completely free, licensed under MIT-0. You can download, install and use it at no cost.
Which platforms does Context Switch Token Optimizer support?
Context Switch Token Optimizer is cross-platform and runs anywhere OpenClaw / Claude Code is available (cross-platform).
Who created Context Switch Token Optimizer?
It is built and maintained by nicochow (@nicochow); the current version is v1.0.1.
More Skills