skulptor agent session
U

Add dark mode toggle to settings page

A

I'll search for the settings component and existing theme utilities first.

T

Grep: Found ThemeContext in src/contexts/theme.tsx

A

Found it. Adding toggle with your existing theme system.

T

Edit: src/components/Settings.tsx (lines 42-58)

S

Score: A (Tests: pass, Lint: clean)

AI Agents

AI That Thinks With Your Codebase, Not Just Your Prompt

Connect any provider. Deploy specialized agent fleets. Every action is scored, logged, and reversible. Your codebase becomes the context, not an afterthought.

ClaudeGPTOllamaCustom Endpoints

An Agent Fleet, Not a Single Bot

Multiple specialized agents work in parallel on different tasks. Each agent has its own context window, tool permissions, and focus area.

Architect Agent

Plans structure, manages dependencies, creates scaffolding

Code Agent

Writes, edits, and refactors code across files with full AST awareness

Review Agent

Runs tests, checks lint, validates changes before commit

Built-in Tool System

Agents don't just generate text. They execute real operations on your codebase through a typed, permissioned tool system.

Read

Read any file

Write

Create files

Edit

Precise edits

Bash

Run commands

Glob/Grep

Search code

Git

Version control

Model Context Protocol

MCP: Connect to Any Data Source

7 built-in MCP servers ship with Skulptor, and a community marketplace lets you add more. Give your agents access to databases, APIs, documentation, and any external system.

Filesystem access with permission boundaries
Database querying (PostgreSQL, MySQL, SQLite)
GitHub integration (PRs, issues, actions)
Documentation search and retrieval
Browser automation for testing
Custom REST/GraphQL endpoint connectors
Community server marketplace
.skulptor/mcp.json
{
"servers": {
"github": { "command": "mcp-github" },
"postgres": { "command": "mcp-pg" },
"browser": { "command": "mcp-puppeteer" }
}
}

Every AI Action, Scored A through F

Rollback scoring evaluates every change across five dimensions. Bad scores trigger automatic rollback before your codebase is affected.

Overall Score

A(94/100)

Status

Changes Applied
Tests
A30%
Lint
A25%
Health
B20%
Intent
A15%
Breaking
A10%

Context Engine

Agents understand your entire repository through tree-sitter parsing, conversation compression, and project-level rules.

Tree-sitter Repo Map

Every symbol, function, class, and import is indexed into a navigable map. Agents know where everything lives.

Conversation Compression

Long sessions are intelligently compressed so agents retain critical context without hitting token limits.

.skulptor/rules.md

Project-level instructions that every agent reads on startup. Define coding standards, naming conventions, and architecture decisions.

Auto Changelog and Session Files

Every agent session produces a structured log. Changes, decisions, errors, and rollbacks are all documented automatically so your team always knows what happened.

/docs/sessions/04-06-26.md