What Bootstrapping Does
On each Agent invocation, Zeus performs the following bootstrapping flow:- Assemble System Prompt — Concatenate core prompts from multiple Bootstrap files in order
- Select Mode Prompt — Inject corresponding behavioral guidelines based on current mode (Agent / Ask / Plan)
- Dynamic Context Injection — Inject current time, user/project profile, relevant memories, Skills metadata, etc.
- Load Tool Set — Filter by mode and assemble available tools
Bootstrap Files
Zeus’s System Prompt uses a layered architecture composed of multiple Markdown files. Each file handles an independent responsibility and is concatenated in order at startup.Core Files
| File | Responsibility | Content Summary |
|---|---|---|
CORE.md | Core identity & capability declaration | Programming languages, information processing, content creation, file handling, limitations |
SOUL.md | Personality & values | Professional text style (no emoji), accuracy, privacy, security principles |
TOOLS.md | Tool usage guide | Five-layer tool hierarchy description and selection rules, dynamically replaces current available tool list |
WORKFLOW.md | Task processing flow | Quick response vs complex task, task planning rules, completion criteria |
MEMORY.md | Memory system instructions | Trigger words, tool usage, type selection, scope, priority rules |
System Prompt Full Description
View detailed content and design principles for each Bootstrap file
Mode Prompts
Different behavioral prompts are injected based on the current interaction mode:| File | Mode | Behavior |
|---|---|---|
agent_mode.md | Agent | Full tool access, proactive execution and modification |
ask_mode.md | Ask | Read-only tools, analysis and Q&A, guides mode switching |
plan_mode.md | Plan | Read-only (no execution), research and planning, outputs structured plans |
invoke() entry point. In Ask mode, write tools are replaced with placeholders (progressive disclosure) — the Agent knows the tools exist but cannot call them, and guides the user to switch modes.
Dynamic Injection
On each run, the following context is dynamically injected into the System Prompt:| Injection Item | Description | Source |
|---|---|---|
| Current time | Lets the Agent perceive time | System clock |
| User profile | Summary of user preferences, skills, habits | Memory Profile |
| Project profile | Current project’s tech stack, constraints | Memory Profile |
| Relevant memories | Historical memories related to current conversation | Memory Search |
| Skills metadata | Available skill names and descriptions (without full content) | SkillsManager |
| Connector Skills | Browser / Desktop control instructions | Injected based on connection status |
| MCP prompts | User-selected MCP Prompt templates | MCP servers |
| Resource files | Context files referenced by user via @ | Frontend request |
| Sandbox files | List of existing files in sandbox | Sandbox state |
| Chat attachments | User-uploaded images, documents | Frontend request |
Skills Progressive Disclosure
Skills use a two-phase loading strategy to reduce initial token consumption:| Phase | Content | Token Cost |
|---|---|---|
| Discovery (startup) | Skill name + one-line description | Very low |
| Activation (on-demand) | Full SKILL.md content | On demand |
| Execution (on-demand) | Script files + reference materials | On demand |
load_skill tool.
Skills
Skills definition, management API, and built-in skill documentation
Context
Complete context assembly flow and Token optimization strategies
Memory
Four-layer memory model, user profile & project profile
Tools
Four-layer tool system and mode filtering rules
Related Docs
- Agent Loop — After Bootstrapping completes, enters the execution loop: Agent Loop
- Workspace — Workspace initialization and file storage: Agent Runtime