HOL Guard vs SalesTouch
Autocomplete
HOL Guard is an Apache-2.0, local-first security tool for AI coding agents. On supported integrations it evaluates agent actions including shell commands, secret access, MCP changes and plugin or skill installs, then allows, asks or blocks based on policy. It records decisions locally, with optional cloud sync. Coverage depends on the supported harness and action surface; it is not a network firewall.
Manual
The LinkedIn MCP for AI agents. Connect Claude, Codex, and other MCP-compatible agents to LinkedIn prospect research and automated outreach. Find prospects, research profiles and companies, manage contextual conversations, and follow up with approval controls, scoped OAuth, pacing limits, and activity visibility.
Summary
- HOL Guard scores 1.6/5 and SalesTouch scores 0.8/5 on the Dark Factory Score, a difference of 0.8.
- The widest gap is Integration: HOL Guard 3.0, SalesTouch 1.0.
- HOL Guard scores higher on code quality, integration, self-correction and project complexity.
- They score the same on autonomy.
- HOL Guard is listed as open source and SalesTouch as paid.
- Both are listed under Orchestration.
Dimension by dimension
| Dimension | Weight | HOL Guard | SalesTouch |
|---|---|---|---|
| Autonomy | 28% | 1.5 | 1.5 |
| Code Quality | 20% | 0.5 | 0.0 |
| Integration | 17% | 3.0 | 1.0 |
| Self-Correction | 15% | 1.0 | 0.5 |
| Project Complexity | 20% | 2.0 | 0.5 |
| Dark Factory Score | 1.6 | 0.8 |
Why HOL Guard scores 1.6
HOL Guard is not a code-generating or autonomous-coding tool — it's a policy/gatekeeper layer that intercepts and arbitrates risky actions (shell commands, secret access, MCP/plugin changes) taken by other AI coding agents. Because it does not write, test, or ship code itself, it scores very low on code quality and self-correction, and low on autonomy: its 'decisions' are allow/ask/block policy evaluations rather than independent multi-step task execution. It has no role in generating, refactoring, or iterating on code output, so those dimensions are effectively out of scope for its function. Where it does have some relevance is integration — it explicitly plugs into agent harnesses (Codex, Claude Code, MCP ecosystems) and keeps a local/cloud-synced audit log of decisions, which is a meaningful form of tooling integration, though it's not a CI/CD or deployment pipeline tool per se. On project complexity, since it operates at the action/harness level rather than the codebase-structure level, its applicability is roughly flat across tiers: it can monitor a one-off script's shell commands just as easily as a multi-service app's MCP calls, but it doesn't meaningfully scale its value or sophistication with project size — it offers the same category of oversight regardless of tier, with reduced confidence at true distributed/multi-repo scale where action surfaces multiply beyond a single harness's visibility. Overall, HOL Guard should be understood as a security/governance complement to coding agents rather than a competitor in code generation or autonomy — its Dark Factory Score is necessarily low across the core 'build' dimensions, with modest credit for integration and consistent (if shallow) support across complexity tiers.
Scored by Claude AI — last evaluated October 29, 2026
Why SalesTouch scores 0.8
SalesTouch is not a software-development tool in the sense this rubric targets — it's an MCP connector that exposes LinkedIn prospecting and outreach actions (search, read, message, follow up, publish) to AI agents like Claude or Codex, gated by OAuth scopes, pacing limits, and human approval controls. It has no stated capability to write, test, or ship code, so code quality, CI/CD integration, and project-complexity handling (which are defined here in terms of scripts, modules, and applications) are essentially inapplicable — scores reflect that near-total mismatch rather than poor execution of a dev task. On autonomy, the tool explicitly emphasizes 'approval controls' and 'activity visibility,' meaning actions are designed to be human-gated rather than fully autonomous — appropriate for compliance/safety in outreach, but it caps autonomy scoring under a dev-tool rubric. Integration gets a small credit because it plugs into agent orchestration frameworks (MCP, Claude, Codex) via OAuth, which is a real technical integration pattern, but there's no evidence of git, deployment pipelines, or monitoring hooks relevant to software engineering. Self-correction is unaddressed in any materials (no mention of retry logic, error recovery, or iterative refinement), and project-complexity tiers (script through distributed system) don't meaningfully apply since this tool operates in the sales/outreach domain, not software construction — so the tier scores are flat and minimal across the board.
Scored by Claude AI — last evaluated October 22, 2026
More comparisons
Generated from Dark Factory Score data. How the score works. Comparing three or more tools at once is part of Pro.