Quick Start
AI coding harness in about 5 minutes — structural scan, agent briefings, and verify gates. No repo.sruja required on day one.
Step 1: Install the CLI (1 minute)
curl -fsSL https://sruja.ai/install.sh | bash
sruja start -r .
sruja drift -r . --structural-only --advisory
You get file-level findings (cycles, layer violations, hub modules) and a could not infer section where the scan is uncertain.
Step 2: Wire your editor (MCP + harness skill)
MCP (Cursor template in the repo): register sruja mcp with profile coding and SRUJA_MCP_READONLY=1. See Host agent integration.
Skills:
npx skills add https://github.com/sruja-ai/sruja --skill sruja-harness
Optional — reviewed architecture in Git:
npx skills add https://github.com/sruja-ai/sruja --skill sruja-architecture
Host loop:
focus / drift → your agent edits code → verify-task → (optional) agent record on failure
Step 3: Run gates locally
# Before a task (paste briefing into the agent)
sruja focus -r . --file path/to/file.rs
# After the agent edits
sruja verify-task --profile coding -r .
In VS Code/Cursor: Sruja: Brief this file, Sruja: Verify task, Sruja: Register MCP Server.
Step 4 (optional) — Reviewed repo.sruja
When you want versioned architecture intent in CI, ask your agent:
Use sruja-architecture. Gather evidence, generate or update repo.sruja,
then sruja lint repo.sruja and sruja drift -r . -a repo.sruja.
Diagrams are exports, not the product center:
sruja export mermaid repo.sruja > architecture.mmd
Quick Reference
| What you want | Command |
|---|---|
| First scan | sruja start -r . then sruja drift -r . --structural-only --advisory |
| Harness skill | npx skills add … --skill sruja-harness |
| Brief agent | sruja focus -r . --file <path> |
| After AI edit | sruja verify-task --profile coding -r . |
| Architecture skill | npx skills add … --skill sruja-architecture |
| Lint reviewed truth | sruja lint repo.sruja |
Common Questions
"What if sruja isn't found?"
export PATH="$HOME/.local/bin:$PATH"
# or: just build (or make build) && use target/release/sruja
"Do I need MCP?"
No for CLI-only workflows. MCP is the best experience inside Cursor/Claude Desktop.
"Is Sruja a coding agent?"
No. Cursor/Copilot own the LLM loop. Sruja is the harness (drift, focus, verify-task).
"What's the difference between quickstart and discover?"
quickstart/start— human-friendly onboardingdiscover— detailed JSON for AI generation (architecture skill)
VS Code extension
Install Sruja Language Support for .sruja syntax, validation, harness commands, and MCP registration.