Tools
Core tool catalog, visibility rules, and extension surfaces.
AlphaBase tools are callable actions available to the agent loop.
Tool Sources
Tool source types in runtime:
builtinmcppluginlocal-jsendpointopenapi-generated
Core Built-In Tool IDs
The built-in runtime catalog includes:
- Filesystem/read:
read,list,glob,grep,codesearch,search_history - Filesystem/write:
write,edit,apply_patch - Shell:
bash - Web:
webfetch,websearch,websurfer - Delegation:
task,task_send,task_wait,task_stop,task_list,task_inspect,task_kill - Planning/task state:
todoread,todowrite - Interactive:
question - Skill/knowledge:
skill,lessons - Diagnostics/system:
provider_health,session_abort,cerberus_scan,troubleshoot,context_usage,cost_dashboard,mcp_status,mcp_reconnect,checkpoint_manage,git_status,settings_read,settings_update
Additional tools can appear based on flags/config (for example lsp, batch, plan tools).
Visibility Gating
Runtime visibility is context-aware:
codesearchis shown for AlphaBase provider sessions or whenALPHABASE_ENABLE_EXAis enabledwebsurferis shown for AlphaBase provider sessions or whenALPHABASE_ENABLE_WEBSURFERis enabledwebsearchis part of core catalog and still permission-gated at execution- coworker-prefixed endpoint tools are hidden outside coworker contexts by default
Permission Mapping
Tool execution is controlled by Permissions.
Important mapping:
editpermission governs file-mutation behavior (edit,write, patch/multiedit style operations,apply_patchworkflow)
Tool Metadata Model
Tools can publish structured metadata used in governance and UI, including:
- capability tags
- recommended policy (
allow/ask/deny) - risk level
- tenant scope
- auth requirements
Extension Surfaces
You can extend tooling via:
- Custom Tools (
.alphabase/tools/*.ts|js) - Plugins
- MCP Servers
- Endpoint tools (tools-hub adapters)
Example Policy
{ "permission": { "read": "allow", "edit": "ask", "bash": { "*": "ask", "git status*": "allow", "git diff*": "allow" }, "webfetch": "allow", "websearch": "allow" }}