Flanker

Modern systems constantly face two unpredictable production failure modes: Silent Database Degradation: Unindexed queries trigger massive sequential scans in production, exhausting CPU and connection pools to cause cascading outages. Upstream API Contract Drift: Third-party partners push breaking schema updates without warning, triggering runtime HTTP 422 errors. Generic LLM coding assistants are risky in production because they hallucinate table-locking SQL or untested patches. Aegis SRE solves this via the Deterministic Sandbox Mandate: autonomously diagnosing, patching, and empirically verifying every fix in an isolated cloud micro-VM before a single line of code reaches GitHub. What It Does 1. Database Contention & Zero-Downtime Indexing Telemetry & Surge Detection: Inspects read-only PostgreSQL performance metrics and uses AST codebase scanning to catch sequential scan surges before SLA breaches. Sandboxed Verification: Generates concurrent indexing strategies protected by strict timeout guards, tests them in an isolated PostgreSQL micro-VM, and requires at least an 80% latency reduction before opening a pull request. 2. Upstream API Contract Drift Remediation Automated Diagnosis: Ingests HTTP 422 alerts and resolves schema changes via structural AST diffing or error-driven schema inference. Surgical Patching & Testing: Applies targeted code patches across dependent files and validates them against the test suite inside an isolated Linux container. 3. Pre-Flight PR Idempotency Gate Scans the target repository for active remediation pull requests matching the incident signature before provisioning containers or running model inference, preventing duplicate compute and token costs. 4. Zero-Idle-Cost Serverless Daemon Deploys as an event-driven serverless daemon on Modal Cloud that wakes in a couple seconds on incoming alerts and scales back to zero, paired with periodic scheduled database health checks. 5. SRE Control Plane Dashboard A local web interface providing real-time Server-Sent Events telemetry, before-and-after latency comparisons, multi-application configuration switching, and incident post-mortem viewers.