What SecHelix does — and what it refuses to pretend.
Short, factual answers for engineers, security teams, evaluators, and AI/search systems trying to understand the project.
01
Frequently asked questions
What is SecHelix?SecHelix is an open-source, evidence-first application-security Agent Skill and orchestration methodology for repositories and environments you are authorized to test. It maps attack surfaces and trust boundaries, selects applicable security hypotheses, coordinates specialist review, independently verifies material findings, fixes root causes, requires regression proof, and produces a release decision.Is SecHelix a vulnerability scanner?Not by itself. SecHelix can normalize evidence from tools such as Semgrep, CodeQL, OSV, Gitleaks, Trivy, Playwright, ZAP, Nuclei, and package audits, but scanner alerts remain hypotheses until SecHelix establishes attacker control, reachability, a failed boundary, concrete impact, root cause, and proof.How do I install SecHelix?The recommended install command is: npx skills@latest add omarmohelal/SecHelix --skill sechelix. The public repository also includes project-local adapters for Claude Code, Codex, GitHub Copilot/Agent Skills-compatible clients, and a vendor-neutral portable skills bundle.Which security areas does SecHelix cover?The current catalog contains 546 structured hypotheses across authentication, sessions, authorization and BOLA/BFLA, injection, API security, files and uploads, SSRF, browser/client security, business logic, payments/accounting, race conditions and idempotency, databases, cryptography and secrets, supply chain, CI/CD, cloud/configuration, privacy/logging, AI/Agent/MCP, operational security, release security, and attack-surface mapping.How does SecHelix reduce false positives?A scanner alert or model suspicion does not become a trusted finding automatically. High and Critical candidates require an independent verification/refutation pass, and FALSE_POSITIVE, UNKNOWN, and BLOCKED are valid retained outcomes when the evidence does not support a vulnerability claim.Can SecHelix audit AI agents and MCP servers?Yes. When an application actually uses AI/LLMs, RAG, agents, tools, memory, or MCP servers, SecHelix can model those trust boundaries and assess prompt injection, unsafe tool authorization, cross-user memory or retrieval leakage, poisoning, unsafe model output reaching security-sensitive sinks, SSRF through tools, excessive agency, and related supply-chain risks.Can a company use SecHelix with existing security tools?Yes. SecHelix is designed to sit above existing SAST, SCA, DAST, browser, runtime, and organization-specific checks. It does not require replacing existing scanners; it normalizes evidence and applies one verification and release-decision standard.Does SecHelix publish measured accuracy or benchmark scores?One, narrowly. The blind label suite has a single uncontaminated run (2026-09-02): precision 0.950, detection recall 1.000, false-positive rate 0.053 across 76 authored cases. It is label-only — one question per file — so it does not measure the independent verifier, the adapters, remediation, regression proof or the release gate. Full-workflow metrics including applicability accuracy, regression-proof rate and release-gate accuracy remain NOT_MEASURED, and 0.950 must not be quoted as SecHelix accuracy.Can SecHelix be used on production systems?Only in a bounded PRODUCTION_SAFE mode for explicitly authorized systems, using non-destructive evidence and verification. Tests that could mutate money, identity, inventory, customer data, external-provider state, or availability belong in local or staging fixtures unless separately approved.What release decisions can SecHelix produce?The release contract uses PASS, PASS_WITH_KNOWN_RISK, BLOCKED, and INCOMPLETE. Missing required evidence fails closed; UNKNOWN and BLOCKED checks are not silently converted into NOT_APPLICABLE.