Skip to content

OSS Artifact Fleet Gap Analysis

Baseline before the fleet

SecOpsAI already had ecosystem-specific artifact analysis, crates build.rs rules, package registry watchers, research cases, a bounded model bridge, and review-only Blog Ops. It did not have one artifact-level queue joining those capabilities, a central rule-pack contract, a staged triage record, or verified high-volume throughput metrics.

Implemented coverage

Gap Current state
Metadata-only stage Cursor/index records and dedupe database exist; live collectors are bridged where available
Deterministic stage Safe archive inspection, ecosystem rules, generic rule pack, IOC/hash evidence, and Rust/build-hook rules exist
Model stage Minimized context builder and explicit triage record exist; no model call occurs for clean artifacts
Analyst stage Suspicious/inconclusive statuses and analyst queue exist
Rust campaign Safe proc-macro1, proc-macro-en, and legitimate proc-macro2 fixtures and regression tests exist
Research handoff Review-only artifact handoff and Blog Ops draft command exist
Dashboard Artifact Fleet status, queue counts, and degraded/not-configured state are visible under Enterprise Security

Remaining limitations

  • The current local implementation is not proof of a 114,000-artifact/day production fleet.
  • Some ecosystems have global collectors through the existing research worker; others still require a dedicated metadata adapter or authorized marketplace integration.
  • The generic YARA pack has structural validation when yara-python is absent; production deployments should enable native compilation and rule telemetry.
  • Binary analysis is metadata and safe-string oriented; it is not a sandbox or full reverse-engineering system.
  • Model calls require an available configured bridge/provider and remain approval-gated at the analyst and publication boundaries.
  • Cost accounting becomes meaningful only after a worker records provider token usage and price configuration.

These limitations are surfaced as explicit status values rather than being flattened into a clean result.