All scenes
Agent Implementation
AI Agent Coding
forge-agentfeat/tool-approval
PROMPT · 0%
AGENT REASONING
Define the trust boundaryReads may be automatic, while every write must be explicit and contained.
Explicit approval required
Path contained to workspace root
6const workspaceTools = {
7 readFile: tool({
8 description: 'Read a UTF-8 file inside the active workspace',
9 inputSchema: z.object({ path: z.string().min(1) }),
10 execute: ({ path }) => readWorkspaceFile(path),
11 }),
12 applyPatch: tool({
13 description: 'Apply a reviewed patch to the active workspace',
14 inputSchema: z.object({
15 path: z.string().min(1),
16 patch: z.string().min(1),
17 approved: z.literal(true),
TOOL APPROVALAwaiting verification
CHECKAbout this scene
AI Agent Coding
An AI coding agent inspects a repository, wires a guarded tool, edits the runtime, and verifies the new behavior across synchronized panels.
Designed for deep work, streaming backgrounds, and the quiet pleasure of watching software take shape.