Version: 2026-06-05
Summary
- Vendor review is not only a privacy topic; it also covers supply chain, updates, plugins, rule packs, and tool rights.
- Large vendors may be easier to approve, but not trusted blindly.
- Small tools, plugins, MCP servers, rule packs, and skills are reviewed more restrictively.
- Experiments are allowed only in isolation and not on machines with access to customer projects.
- Security patches should be applied quickly; new capabilities are reviewed deliberately.
- Latest is not a production update strategy.
- Tools without clear origin, limited rights, context exclusions, or rollback are not allowed.
Topic
AI vendors, ready-made AI solutions, plugins, local stacks, MCP servers, skills, rule packs, IDE integrations, browser extensions, and package-managed helper tools.
Starting Point
AI ecosystems move quickly. Useful tools appear every week: coding agents, MCP servers, browser extensions, data connectors, prompt packs, editor plugins, local model stacks, hosted model providers, and workflow automation tools.
Speed is useful for experiments, but customer projects need slower decisions.
Risk
A tool may gain access to project code, customer data, local files, issue trackers, repository hosting, browser sessions, shell commands, or CI/CD systems. Vendor claims alone do not prove that this is safe.
Recommendation
Separate security patches from new capabilities. Applying a patch to an approved tool is different from enabling a new connector, plugin, agent mode, or external service.
Evaluate vendors by origin, maintenance, update behavior, data processing, logging, rights, integrations, local footprint, uninstall path, and operational rollback.
Experiments belong in isolated environments without customer project access.
Review Checklist
- Who provides and maintains the tool?
- What data can it see?
- Which systems can it write to?
- How are updates delivered?
- Does it add plugins, rules, or commands?
- Can it be disabled or rolled back quickly?
- Are vendor claims backed by controls the project can verify?
Wall rules
Security patches quickly, new capabilities controlled.
Latest is not a production strategy.
Experiments yes, but not on customer project machines.
Large vendors are easier to review, small tools more restrictive.
Vendor claims are not controls.
Open questions for a project
- Which vendors are already approved?
- Which tool categories are experimental only?
- What is the update policy?
- Who approves new capabilities?