davidl/tolaria
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

feat: enable full shell access for AI agent + simplify MCP tools

Browse Source 
Remove --tools "" restriction so the agent has native bash/read/write/edit
access. Set vault path as working directory for the subprocess.

Simplify MCP to 4 Laputa-specific tools (search_notes, get_vault_context,
get_note, open_note) — everything else is handled by native tools.

Add file operation detection from Write/Edit tool calls to auto-open
created notes and refresh modified notes in the UI. Enhanced tool call
labels show bash commands, file paths, and note names.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
This commit is contained in:
Test
2026-03-05 12:12:12 +01:00
parent 6a57e83c99
commit 89da970455
12 changed files with 311 additions and 420 deletions
Download Patch File Download Diff File Expand all files Collapse all files

5
src/utils/ai-agent.test.ts
View File

@@ -12,13 +12,14 @@ import { buildAgentSystemPrompt, streamClaudeAgent } from './ai-agent'
describe('buildAgentSystemPrompt', () => {
it('returns preamble when no vault context', () => {
const prompt = buildAgentSystemPrompt()
expect(prompt).toContain('AI assistant integrated into Laputa')
expect(prompt).toContain('working inside Laputa')
expect(prompt).toContain('full shell access')
expect(prompt).not.toContain('Vault context')
})
it('appends vault context when provided', () => {
const prompt = buildAgentSystemPrompt('Recent notes: foo, bar')
expect(prompt).toContain('AI assistant integrated into Laputa')
expect(prompt).toContain('working inside Laputa')
expect(prompt).toContain('Vault context:')
expect(prompt).toContain('Recent notes: foo, bar')
})

19
src/utils/ai-agent.ts
View File

@@ -1,7 +1,8 @@
/**
* AI Agent utilities — Claude CLI agent mode with MCP vault tools.
* AI Agent utilities — Claude CLI agent mode with full shell access + MCP vault tools.
*
* The Claude CLI handles the tool-use loop internally via MCP.
* The agent has full native tool access (bash, read, write, edit) plus
* Laputa-specific MCP tools (search_notes, get_vault_context, get_note, open_note).
* The frontend receives streaming events for text, tool calls, and completion.
*/
@@ -9,10 +10,14 @@ import { isTauri } from '../mock-tauri'
// --- Agent system prompt ---
const AGENT_SYSTEM_PREAMBLE = `You are an AI assistant integrated into Laputa, a personal knowledge management app.
You can perform actions on the user's vault using the provided tools.
Be concise and helpful. When creating notes, use appropriate entity types and folder conventions.
When you've completed a task, briefly summarize what you did.`
const AGENT_SYSTEM_PREAMBLE = `You are working inside Laputa, a personal knowledge management app.
Notes are markdown files with YAML frontmatter. Standard fields: title, type (aliased is_a), date, tags.
You have full shell access. Use bash for file operations, search, bulk edits.
Use the provided MCP tools for: full-text search (search_notes), vault orientation (get_vault_context), parsed note reading (get_note), and opening notes in the UI (open_note).
When you create or edit a note, call open_note(path) so the user sees it in Laputa.
Be concise and helpful. When you've completed a task, briefly summarize what you did.`
export function buildAgentSystemPrompt(vaultContext?: string): string {
if (!vaultContext) return AGENT_SYSTEM_PREAMBLE
@@ -41,7 +46,7 @@ export interface AgentStreamCallbacks {
}
/**
* Stream an agent task through the Claude CLI subprocess with MCP tools.
* Stream an agent task through the Claude CLI subprocess with full tool access.
* The CLI handles the tool-use loop; we receive events for UI updates.
*/
export async function streamClaudeAgent(