Agent security is becoming a runtime stack
A seven-month GitHub, package, and incident-evidence cohort measures identity, sandboxing, authorization, audit, and recovery around autonomous actions.
Supported by the targeted cohort
255 subject-filtered implementations show a real runtime-security category. Sandboxing appears in 37.3% of the cohort, authorization and policy in 13.3%, and runtime identity in 8.6%. Recovery and containment remain thinner at 2.7%, so prevention is forming faster than safe failure recovery.
Market snapshot
Comparable measurements from independent market surfaces.
Implementations and packages
Deduplicated and subject-filtered primary cohort.
Median repository stars
Calculated across repositories with at least one star.
Registry packages
Real npm and PyPI package records, not synthetic entries.
Latest-month npm downloads
Usage surface only; downloads are not equivalent to customers.
Search-match dynamics
Bars show monthly GitHub Search API matches across three runtime-security queries; loaded counts are deduplicated repositories retained for detailed analysis.
What exists inside the category
One item may contain more than one feature.
Runtime identity and credentials
22 · 8.6%
Sandboxing and isolation
95 · 37.3%
Authorization and policy
34 · 13.3%
Audit and approval
33 · 12.9%
Recovery and containment
7 · 2.7%
Monitoring and detection
45 · 17.6%
Representative projects
Cross-source validation
These publications are not part of the primary numeric cohort.