xiage-openclaw-workspace
Summary
A personal OpenClaw workspace powering an AI agent named "虾哥" (Xiage) with structured operational rules, layered memory management, and a curated set of skills for web research, automation, and multi-platform control.
Included Assets
Core Documents
SOUL.md— Agent identity and personality rules (direct, concise, no corporate speak)AGENTS.md— Operational manual: startup rules, memory layering, tool call lock mechanismMEMORY.md— Long-term memory index (hot/warm/cold layer entry points)TOOLS.md/docs/core/TOOLS.md— Local environment notes and API key config (includes Tavily key)package.json— Workspace dependencies:@base44/sdk,pdf-parse,tesseract.js
Skills (15 total)
| Skill | Purpose |
|---|---|
joeseesun-opencli-skill | CLI control for Bilibili, Zhihu, Twitter/X, YouTube, Weibo, Reddit, and 12 more platforms via Chrome session reuse |
Shaivpidadi-free-ride | Auto-manages OpenRouter free models to avoid rate limits |
matrixy-Agent Browser | Browser automation with Chrome debug mode |
NicholasSpisak-clawddocs | Documentation expert with decision tree navigation and search |
ide-rea-Baidu Search | Baidu search integration |
gpyangyoujun-Multi Search Engine | Multi-engine search |
halthelobster-Proactive Agent | Proactive agent behaviors |
ivangdavila-Self-Improving + Proactive Agent | Self-improvement and proactivity |
jk-0001-Automation Workflows | Automation workflow tooling |
byungkyu-API Gateway | API Gateway skill |
chindden-Skill Creator | Skill creation tooling |
Plus 4 additional skills with package.json config only.
How to Use
Startup sequence (per AGENTS.md):
- Read
SOUL.md→ agent identity - Read
USER.md→ current user context - Read
memory/hot/current.mdandmemory/hot/lessons.md→ hot memory - Read
memory/YYYY-MM-DD.md(today + yesterday) → daily logs - Read
MEMORY.md→ long-term memory index
Memory layers:
- Hot (
memory/hot/) — current state, lessons; loaded every session - Warm (
memory/YYYY-MM-DD.md) — daily logs; loaded today + yesterday - Cold (
memory/cold/archive.md) — historical decisions; loaded on demand - Index (
MEMORY.md) — entry point for all layers
Web reading priority (per TOOLS.md): opencli → agent-browser → web_fetch/summarize
Document write lock: All protected doc edits require explicit "同意变更并升级版本" (consent and version bump) from the user before execution.
Notes
- The agent enforces a strict no-corporate-speak rule: be direct, concise, and opinionated.
- Web scraping for Chinese platforms (Zhihu, WeChat) has platform-specific strategies documented in
TOOLS.md. opencli-skillrequires an active Chrome session for authentication — no API keys needed for supported platforms.free-rideis useful for reducing OpenRouter API costs by automatically cycling free models.
No comments yet.