Evenfire is an intelligence layer platform for running LLM agents that take real actions, on your infrastructure
TypeScript
export type CodeDefault = 'required' | 'skip'
export interface NativeToolMeta {
name: string
/**
* Resolved value of the tool class's hard-coded `requiresApproval()` in
* mcp-host. Verified against `mcp-host/src/core/tools/*.ts` —
* - `'required'` (true): http_request, shell_exec
* - `'skip'` (false): everything else, including all clerum__* output
* tools (registered via InternalToolAdapter) and the file/json/system
* tools.
* The `isRisky` warning predicate fires only when the operator overrides
* a `'required'`-default tool to Skip — that is the only direction that
* actually loosens approval. Setting any tool to "Required" tightens
* (or no-ops if the default is already required).
*/
codeDefault: CodeDefault
description: string
/**
* Operator-facing tooltip shown next to the warning icon when the override
* loosens approval (state === 'skip' && codeDefault === 'required').
* Only meaningful for tools whose code default is `'required'`.
*/
riskHint?: string
}
export const NATIVE_TOOLS: NativeToolMeta[] = [
{
name: 'clerum__generate_docx',
codeDefault: 'skip',
description: 'Generate .docx output from agent steps.',
},
{
name: 'clerum__generate_markdown',
codeDefault: 'skip',
description: 'Generate Markdown output from agent steps.',
},
{
name: 'clerum__generate_pdf',
codeDefault: 'skip',
description: 'Generate PDF output from agent steps.',
},
{
name: 'clerum__generate_xlsx',
codeDefault: 'skip',
description: 'Generate .xlsx output from agent steps.',
},
{
name: 'clerum__list_workflows',
codeDefault: 'skip',
description: 'List available marketplace recipes.',
},
{
name: 'clerum__trigger_workflow',
codeDefault: 'skip',
description: 'Trigger a plugin by name.',
},
{
name: 'file_read',
codeDefault: 'skip',
description: 'Read files in the agent workspace.',
},
{
name: 'file_write',
codeDefault: 'skip',
description: 'Write files in the agent workspace.',
},
{
name: 'http_request',
codeDefault: 'required',
description: 'Outbound HTTP requests (GET/POST/PUT/DELETE).',
riskHint:
'Skipping approval relies on CLERUM_HTTP_ALLOWLIST being set in the mcp-host ConfigMap. Confirm that gate before applying.',
},
{
name: 'json_transform',
codeDefault: 'skip',
description: 'Transform JSON via JMESPath / jq-style queries.',
},
{
name: 'shell_exec',
codeDefault: 'required',
description: 'Execute shell commands in the agent workspace.',
riskHint:
'Skipping approval allows arbitrary command execution. There is no allowlist analog for shell commands.',
},
{
name: 'system_info',
codeDefault: 'skip',
description: 'Read OS / process metadata (CPU, memory, env).',
},
]