// Security disclosures & positioning

Report a vulnerability.
Or read why we don't replace your security stack.

Two things on this page. First, how to disclose a security issue in ArgosBrain — channel, scope, response time. Second, an explicit positioning statement we wish more vendors made: ArgosBrain is a memory layer, not a replacement for the tools you already trust. We complement grep, Semgrep, CodeQL, SonarQube, and the rest of your security stack. We are not a substitute for any of them.

How to report a vulnerability

Email [email protected]. PGP encryption is supported on request — reply to your initial mail and we'll send the public key inline.

Please include: a clear description, steps to reproduce, the affected component (binary version, MCP server version, dashboard URL, or web property), and an impact assessment if you have one. If you've already drafted CVE-style metadata, attach it; otherwise we'll work it out together during triage.

PGPon request
Initial responsewithin 72 hours, business days
Triage decisionwithin 7 days of initial response
Fix or mitigationwithin 30 days for High/Critical, 90 days for Medium/Low
Disclosurecoordinated, default 90 days from triage
Acknowledgmentresearchers credited in /updates unless they opt out

Scope

What we want to hear about, what we'd rather you not file.

// In scope
  • The argosbrain, argosbrain-mcp, and argosbrain-manager binaries (any released version)
  • argosbrain.com and app.argosbrain.com (the dashboard, install endpoints, license-token exchange)
  • The local dashboard at 127.0.0.1:3733 (auth, CSRF, XSS, SSRF, RCE, path traversal)
  • Supabase Edge Functions used for license exchange, install tokens, and publish-hash
  • Build pipeline + release artifacts (tarballs, hashes, signing)
  • Any data leakage from the egress promise (source, file paths, query content)
// Out of scope
  • Issues that require physical access to a developer's machine
  • Self-XSS that requires the user to paste attacker-controlled content into devtools
  • Rate-limiting / brute-force on public marketing pages (no auth boundary)
  • Reports about CVEs in transitive dependencies that we cannot reach in our threat model
  • Findings that require a malicious MCP host to be installed first (the host is trusted)
  • Theoretical timing oracles below 50ms on the local socket (not a meaningful attack vector)

How ArgosBrain stacks with the rest of your security tooling

This is the part most vendors skip. We want it on the record. ArgosBrain is not a security scanner. It is a structural memory layer that the rest of your stack queries to make their findings durable, reachable, and queryable across sessions.

Tool What it does What ArgosBrain adds
grep / ripgrepFast literal text search (comments, logs, non-code strings)Structural lookup the same query can't answer — exact callers, blast radius, reachability
SemgrepPattern-based rule matching, AST-awareMemory of where Semgrep findings touched code, surviving refactors and re-runs
CodeQLDeep dataflow + taint analysisReachability cache + cross-session symbol map so the agent doesn't re-derive the call graph every prompt
SonarQubeCode quality, duplications, complexity hotspotsStructural neighborhood for every Sonar finding — what calls it, what it calls, who else touches the cluster
Snyk / DependabotVulnerability scanning of dependenciesTrace from a CVE to every code path that imports the vulnerable symbol — supply-chain reachability
Tree-sitter / SCIP / LSPParsers and indexers we build on top ofIn-memory graph that survives editor restarts and serves agents over MCP at sub-millisecond cost
Cursor / Claude Code / CodexAI coding agents that consume code contextPersistent, deterministic memory layer they query via MCP — instead of re-grepping every turn
Use grep for free-text. Use Semgrep / CodeQL / SonarQube / Snyk for what they were built for. Use ArgosBrain to make their findings live longer than the session.

If you find a security tool we should pair with and don't yet, tell us at [email protected]. The list above is the canonical pairing matrix and we update it as the ecosystem evolves.

Safe-harbor for security researchers

If you make a good-faith effort to comply with this policy, we will:

Good faith means: no DDoS, no privacy invasion of other users, no degradation of service, and no exfiltration of data beyond the minimum necessary to demonstrate the issue. If your research requires touching production data or other users' codebases, stop and email us first.