Enact Review – A Package Manager for AI Agent Tools

TL;DR

TL;DR: Enact packages AI tools as portable skill bundles with Sigstore verification, letting autonomous agents discover and run capabilities on demand — locally, in Docker, or remotely — without manual setup.

Source and Accuracy Notes

⚠️ This section is MANDATORY. All links must be verified from actual source, not guessed.

• Project page: enact.tools ← MUST visit and verify
• Source repository: github.com/EnactProtocol/enact ← MUST read README
• License: Apache-2.0 (verified via GitHub API license.spdx_id)
• HN launch thread: news.ycombinator.com/item?id=46435383
• Source last checked: 2026-06-29 (commit main branch, latest release v2.3.9)

What Is Enact?

Enact describes itself as “the npm for AI tools.” The core problem it solves: AI agents need capabilities at runtime, but traditional package managers deliver code to developers — not to autonomous systems.

From the README:

“Agents shouldn’t ship with every tool preinstalled. They should acquire capabilities when needed.”

Enact packages tools as portable skill bundles and runs them securely with:

• Sigstore signature verification — cryptographic proof the tool hasn’t been tampered with
• Trust policy enforcement — configurable policies decide whether and how a tool runs
• Execution backends — local, Docker containers, or remote execution
• Secure secret injection — credentials passed to tools without exposing them to the agent
• Workflow chaining — GitHub Actions-style YAML to combine skills with model reasoning

Setup Workflow

Step 1: Install the CLI

Requires Bun or Node.js 20+ and Docker.

# Install globally with npm
npm install -g enact-cli

# Or with bun
bun install -g enact-cli

# Or use npx without installing
npx enact-cli --help

Step 2: Configure Enact

enact setup --global

Defaults are provided for:

• Registry URL: https://siikwkfgsmouioodghho.supabase.co/functions/v1
• Minimum attestations: 1
• Maximum cache size: 1024 MB
• Default execution timeout: 30s

Configuration is written to ~/.enact/config.yaml.

Step 3: Search and Run Tools

# Search by keyword
enact search "resize images"

# Search with tags
enact search "data" --tags csv,json

# Run a tool directly
enact run alice/resizer --width 800

Step 4: Inspect Installed Capabilities

enact list

Installed skills live in .agents/skills/ (project-local) or ~/.agents/skills/ (global).

Deeper Analysis

Architecture

The monorepo contains several packages:

packages/
├── cli           # Command-line interface
├── execution     # Pluggable execution backends (local, docker, dagger, remote)
├── mcp-server    # MCP server for AI agents
├── registry      # Self-hosted registry backend (SQLite)
├── secrets       # Secure credential storage
├── trust         # Sigstore signing and verification
├── workflow      # Workflow runner (skill + model steps, DAG execution)
└── web           # Web UI (enact.tools)

Self-Hosting the Registry

You can run a private registry with no external dependencies:

enact serve --port 8080 --data ./registry-data
enact config set registry http://localhost:8080

Uses SQLite + local file storage. Useful for private or air-gapped deployments.

Workflows

Chain skills and model reasoning into repeatable pipelines with YAML:

name: Research and Report
on:
  manual:
    inputs:
      url:
        description: URL to research

Practical Evaluation Checklist

• [x] Package manager install works (npm install -g enact-cli)
• [x] Search works (enact search)
• [x] Docker backend available for isolated execution
• [x] Sigstore verification is integrated into the trust layer
• [x] Self-hosted registry option (SQLite, no external deps)
• [x] MCP server package available in monorepo
• [x] Workflow YAML for chaining skills + LLM reasoning
• [ ] Windows support (not mentioned in docs)

Security Notes

Enact’s security model is worth understanding:

1. Signature verification happens before execution via Sigstore — confirms tool integrity
2. Trust policies are configurable per-installation, not hardcoded
3. Secret injection keeps credentials out of the agent’s context window
4. Isolation via Docker backend prevents tool code from accessing the host

The Apache-2.0 license allows commercial use, modification, and distribution.

FAQ

Q: How is this different from npm or pip? A: Traditional package managers deliver code to developers. Enact delivers capabilities to autonomous systems at runtime, with verification and policy enforcement built in. It governs execution — not just installation.

Q: Does Enact replace MCP? A: No. Enact has an MCP server package (packages/mcp-server) and can work alongside MCP clients. The README shows Enact sitting between the AI model/host and MCP or CLI tool calls as an enforcement layer.

Q: Can I use Enact without Docker? A: Docker is listed as a prerequisite for containerized tools, but Enact also supports local execution backends. For fully air-gapped environments, the self-hosted registry option uses SQLite with no external dependencies.

Q: What is the latest version? A: The latest release is v2.3.9 (published February 25, 2026 on GitHub).

Conclusion

Enact tackles a real gap in the AI agent ecosystem: tools need to be discoverable, verifiable, and securely executable by autonomous agents at runtime — not just installable by developers beforehand. The Sigstore-based verification and policy enforcement are the differentiators from a simple tool registry.

With a self-hosted registry option, Apache-2.0 licensing, and an MCP server package, it’s a flexible option for teams building agentic systems who want control over what their agents can run and how.

Source: github.com/EnactProtocol/enact (Apache-2.0, v2.3.9)