Everything a repository can get wrong, in one place.
Proven open-source engines and Sentinel's own checks, normalised into one finding model, with owners and decisions you can audit.
Secrets and credentials
Keys and tokens anywhere in git history, plus committed .env and credential files. Reported by path and commit, never by value.
Vulnerable dependencies
Known advisories in every tracked lockfile. Development-only packages and unpinned versions are labelled, so a test tool or a guessed version does not read like a production emergency.
Code security
Static analysis with community rules plus rules written from real incidents, such as unescaped JSON in pages and model output rendered as HTML.
AI security
An AI bill of materials per repository: AI SDKs, model identifiers, MCP server configs and agent instruction files. Flags unpinned MCP servers and user input in system prompts.
Next: Agent permissions through rules files, skills and connectors; prompt-injection exposure.
Outdated and end-of-life technology
Container base images and runtimes checked against published end-of-life dates.
Ownership and review
Every repository has an accountable owner; missing CODEOWNERS and unowned code are findings.
SDLC and AI development standards
Unpinned dependencies, missing lockfiles and repository rules such as migration naming limits.
Next: Checks for AI-assisted development practices across the lifecycle.
Privacy and data protection
Next: Personal data in code, logs and AI prompts.
Less noise. More decisions.
One finding model
Ten sources normalised into eight risk categories, each finding with a stable fingerprint that survives rescans.
Owners and risk scores
Every repository has an owner and a 0–100 risk score that counts each issue once, so one advisory in ten lockfiles does not drown out everything else.
Triage with an audit trail
Mark a finding as a false positive or an accepted risk, with a reason and an optional expiry. Decisions are append-only: who, when and why is never overwritten.
Read-only by design
Scanners run in isolated, pinned containers with your code mounted read-only. Sentinel never edits, commits or runs your code.
Context, not alarm
A vulnerability in a test runner is marked development-only. A version guessed from a range is marked as assumed. Rules that only produce false positives on real code are switched off, with the reason recorded.
Organisations and roles
Viewers see findings, triagers record decisions, admins manage members and scanner tokens. Each organisation sees only its own data.
Eight categories, ten sources.
- Secrets and credentials. Keys and tokens anywhere in git history, plus committed .env and credential files. Reported by path and commit, never by value.
- Vulnerable dependencies. Known advisories in every tracked lockfile. Development-only packages and unpinned versions are labelled, so a test tool or a guessed version does not read like a production emergency.
- Code security. Static analysis with community rules plus rules written from real incidents, such as unescaped JSON in pages and model output rendered as HTML.
- AI security. An AI bill of materials per repository: AI SDKs, model identifiers, MCP server configs and agent instruction files. Flags unpinned MCP servers and user input in system prompts.
- Outdated and end-of-life technology. Container base images and runtimes checked against published end-of-life dates.
- Ownership and review. Every repository has an accountable owner; missing CODEOWNERS and unowned code are findings.
- SDLC and AI development standards. Unpinned dependencies, missing lockfiles and repository rules such as migration naming limits.
Where Sentinel is heading.
- Scheduled scans and alerts when a new critical appears
- Deploy gates: block a release on open criticals, with approvals
- AI governance: what each agent may read, write and call
- Token-level AI cost per repository, team and feature
- Semantic and model monitoring for AI features in production
- Code health trends alongside security
- Visibility of your products in AI assistants, measured alongside risk