Deep dives, practical guides, and incident analyses from engineers who build Safeguard. No fluff, no vendor FUD — just what you need to ship secure software.
Install two test MCP servers, one local and one remote, and check whether a tool lists them, their tools and their permissions. A short evaluation test.
Test coding agent guardrails with eight harmless probes: canary files, planted instructions, blocked commands and protected branches. Record pass or fail.
Pin an exact model version in configuration, keep a small evaluation set, and promote a newer version only after it passes. Roll back with one value.
After adoption, sample outputs monthly, track override and reopen rates, rerun a fixed test set, and review vendor change notices. A lightweight monitoring plan.
Inventory agent tooling from four sources: repo configs, developer machines, identity logs and network records. A practical method and what to record.
List model weight files by extension across repos, buckets and registries, then check the format. Pickle based files can run code when loaded. Steps.
Search repos for AI client libraries, API hostnames, keys in config and model loading calls to list the AI services your code uses. A practical plan.
AI coding tools leave config files in repos: CLAUDE.md, AGENTS.md, .mcp.json and .cursor/rules. Search for them to see which agents and servers are in use.
Say what your AI features do, what data they touch, and what you can show. Avoid safe, unbiased, or hallucination-free. A section outline you can adapt.
Weekly insights on software supply chain security, delivered to your inbox.
Your first fix PR is minutes away.
No sales call required, even your agent can complete the purchase over MCP.
