# Valstorm Agent Runtime (vsagent)

> Enterprise-grade autonomous AI execution runtime, ReAct loop, and parallel subagent swarm engine for developer workstations, CI pipelines, and remote team automation.

- **GitHub Repository:** https://github.com/Valstorm-Corp/valstorm-agent
- **License:** PolyForm Noncommercial License 1.0.0 (https://polyformproject.org/licenses/noncommercial/1.0.0)
- **Installation:** `uv tool install --from git+https://github.com/Valstorm-Corp/valstorm-agent.git valstorm-agent --force`
- **Language/Stack:** Python 3.10+ / Rust Native Extensions
- **Web Documentation:** https://www.valstorm.com/free-tools/valstorm-agent

## Key Capabilities
- **Persona-Scoped Tool Registries (-64.8% Tokens):** Instead of dumping 30+ tool schemas into every prompt, vsagent scopes tool access per profile (developer, researcher, archivist, backend-tester), cutting token usage by 64.8% and inference costs by 2.8x.
- **Non-Blocking Parallel Subagent Swarms (background=True):** Spawns isolated child subagents via asyncio.create_task in under 70ms and awaits completion with asyncio.gather(), compressing wall-clock execution time to the slowest single worker.
- **170+ On-Demand Procedural Skills (SKILL.md):** Loads battle-tested standard operating procedures (TDD, bash-scripting, systematic debugging) on demand to eliminate operational regressions like missing chmod +x permissions.
- **Human-in-the-Loop & Ephemeral Sandboxes:** High-risk operations automatically pause via confirmation_required, while untrusted code runs inside isolated Docker containers or E2B sandboxes.

## Quickstart
```bash
# 1. Install vsagent globally via uv
uv tool install --from git+https://github.com/Valstorm-Corp/valstorm-agent.git valstorm-agent --force

# 2. Launch interactive multi-turn REPL with a specialized profile
vsagent chat --profile developer --model gemini-flash-latest

# 3. Execute a headless single-prompt task in CI or local terminal
vsagent chat -p "Run pytest and fix any failing assertions" --profile backend-tester

# 4. Run the 6-axis scientific benchmark & ablation harness
uv run python -m benchmarks.runner --ablation all --mode mock
```