Skip to content

Back to the catalog

Shor
github.comrepo

Shor

Shor is a source-available, multi-tenant autonomous pentesting platform where a ~17-agent LLM pipeline (discovery, hypothesis, adversarial screening jury, exploitation, proof replay) attacks a target URL and optionally its connected source repo. It wields 30+ real offensive tools (sqlmap, dalfox, nuclei, jwt_tool, hydra, ffuf, etc.) plus SAST/secret/SCA tools (Semgrep, Gitleaks, TruffleHog, OSV-Scanner, Trivy) in white-box mode, and only reports findings whose PoC is deterministically re-executed and confirmed by a code-enforced safety layer under a signed Rules of Engagement, with scan authorization gated behind single-use human-minted tokens on its MCP/automation surface.

TL;DR An autonomous multi-agent AI pentesting engine that only reports vulnerabilities it has actually reproduced via a code-enforced, RoE-scoped proof-of-concept safety layer.
Document

What it is

Shor (by Keygraph, Inc.) is a source-available autonomous web pentesting platform. The catalog summary covers the agent pipeline, tool arsenal, and prove-it verification — but omits three things an engineer deciding whether to adopt it needs: the license terms, an opt-in cross-scan memory subsystem, and the specifics of its false-positive machinery.

Licensing (the adoption gate)

Shor is licensed under PolyForm Noncommercial License 1.0.0 — explicitly NOT an OSI open-source license. You may use, modify, fork, and share it for any noncommercial purpose if you keep the copyright/Required Notice lines and credit "Based on Shor by Keygraph, Inc." You may not sell it, bundle it into a paid product, or run it as a paid/hosted service without a separate commercial license. Contributions require signing a Contributor License Agreement. If you intended to offer scanning as a service or embed it in a commercial product, this is a hard stop absent a commercial deal.

Retrieval-memory layer (not in the summary)

Shor ships a two-tier memory system that is feature-flagged and off by default:

  • A per-project tier remembers prior findings for a target so repeat scans build on earlier work.
  • A shared, cross-tenant knowledge tier pre-seeded with distilled MITRE CAPEC and recent web-security write-ups.

Design constraints worth noting: memory feeds agents ideas to try, not verdicts to trust — the prove-it oracle stays the sole arbiter. Third-party research is treated as untrusted input to guard against retrieval poisoning, and promotion of a finding into the shared cross-tenant tier is gated behind consent plus mandatory secret/PII scrubbing. If you enable the shared tier, you are opting into cross-tenant data flow with those guards as your only protection.

False-positive machinery, in detail

Beyond the PoC replay the summary mentions, the end-stage refinement gate scores findings on independent evidence axes (code path confirmed, premise valid, in scope, location verified) that can only lower an over-confident label. Duplicates collapse to one canonical record; out-of-scope or "god-mode" artifacts are demoted, not deleted — everything demoted lands in a review appendix, so nothing is silently dropped.

Outputs and containment specifics

Outputs include SARIF export and fingerprinted findings tracked as new/recurring/resolved across a project's scan history, plus opaque read-only share links and (white-box) per-finding "fix prompts" citing exact file and line. Containment adds detail the summary skips: per-tenant egress firewalling, default-deny scope, per-scan isolated environments with scoped identity/secrets, and cancelling a scan as the kill switch.

Lineage

Shor's discovery+verification pipeline is ported from Anthropic's defending-code reference harness, with agent orchestration following Anthropic's Building Effective Agents (orchestrator-workers, evaluator-optimizer). It cites XBOW, PentestGPT, Shannon, Strix, and HexStrike AI as prior art — useful if you're comparing agentic pentest tools.

Organisations named here

  • MITRE CAPEC
  • PolyForm Noncommercial License 1.0.0

Technologies named here

  • Shor
  • Nmap
  • naabu
  • httpx
  • Katana
  • Ffuf
  • Arjun
  • paramspider
  • subfinder
  • dnsx
  • gau
  • waybackurls
  • kxss
  • wafw00f
  • nuclei
  • sqlmap
  • commix
  • nosqli
  • xsstrike
  • Dalfox
  • ssrfmap
  • sstimap
  • jwt_tool
  • hydra
  • interactsh-client
  • Playwright
  • Semgrep
  • Gitleaks
  • TruffleHog
  • OSV-Scanner
  • Trivy
  • XBOW
  • PentestGPT
  • Shannon
  • Strix
  • HexStrike AI
  • SARIF
  • Model Context Protocol

Techniques named here

  • Hypothesis-driven vulnerability analysis per bug class
  • Adversarial jury screening of candidate findings
  • Automated exploitation with real tools
  • Proof-of-concept replay / deterministic oracle verification
  • Finding refinement and deduplication gate
  • Retrieval-augmented agent memory
  • Black-box scanning
  • White-box scanning