FastClaw / Ability level
Skills and Memory
Skills is a reusable capability package, and Memory is a long-term context layer; together, the two allow the Agent to not only chat, but also to stably execute the process.
Skill model
FastClaw skills uses SKILL.md frontmatter and supports fields such as name, description, homepage, env, metadata.fastclaw/openclaw, etc. Bundled skills include code-runner, data-analysis, fastclaw-api-integration, skill-creator, image-gen, translation, web-search, etc.
- Skills can be installed globally or placed in the Agent's private space.
- Metadata supports emoji, always, OS, requires bins/env/config, install and other descriptions.
- Object storage hydration allows skills to be synchronized between nodes in a multi-copy scenario.
Load priority
The priority described by source code comments is agent workspace > user installed > managed > extra dirs. In other words, Agent private skills can cover lower-level general skills.
| Level | What to put |
|---|---|
| Agent workspace | An Agent’s exclusive workflow and identity binding skills |
| User installed | A user’s own common skills |
| Managed | Official or team skills for unified management of the platform |
| Extra dirs | Read-only mounts, experimental skills, compatible paths |
memory model
README Description MEMORY.md records long-term facts and is updated automatically by heartbeat. Agent context will also reference session history, USER.md and thinking information for subsequent memory retrieval.
Quarantine reminder
When accessing public chat or app_user, USER.md, session, and memory should be partitioned by chatter or end user to avoid injecting facts about one user into another user session.